# SuperCognit — full site content > A no-code platform to build, publish and monetize customer-facing and internal AI agents — trained on your own website and files, with real tool-use and actions, checkout and per-seat billing, and deployment to a website widget, custom domain, WhatsApp and Telegram. This file inlines the full text of every marketing and resource page (https://supercognit.com). Localized copies exist under /pt and /es; this is the English canonical. # Product ## Chat agents https://supercognit.com/product/agents A production chat agent from a prompt Describe what your agent should do, attach your knowledge, and publish. You get a branded, multilingual chat product on your own URL — not a demo, a product your customers and your team actually use. ### What it does - **WYSIWYG behavior** — Edit the prompt and behavior settings visually — tone, guardrails, welcome flow, teaser messages. No YAML, no SDKs. - **Grounded answers** — Agents answer from your knowledge bases with citations, so users can verify every claim against the source. - **Speaks 12 languages** — The chat UI ships localized in 12 languages out of the box, and the agent follows the visitor's language automatically. - **Memory & context** — Agents remember returning visitors and keep long conversations coherent with automatic context management. - **Shows, not just tells** — Ask for a demo or a tour and the agent plays the right video inline — YouTube or Vimeo links from your media library, in our own player with your accent colour and click-to-play, not a pasted embed. Products and options come back as cards. - **Works where your customers are** — Connect a WhatsApp Business number or a Telegram bot and the same agent — same prompt, knowledge and tools — answers there too. Cards and videos adapt to each channel. - **Inbox & contacts built in** — Every conversation lands in your dashboard inbox with visitor metadata — your agent doubles as a lead machine. - **Open or gated access** — Run the agent wide open, require login, or charge for seats — per agent, your call. ### How it works - **Describe it** — Write the prompt in plain language — or start from your website with the one-click importer. - **Attach knowledge & tools** — Connect docs, URLs and tools so the agent can answer accurately and take action. - **Publish** — Share the hosted page, embed the widget, or connect your own domain. Set a price if you want to sell seats. ## MCP servers https://supercognit.com/product/mcp-servers Your expertise, callable from any AI client Package your tools and knowledge as a standards-compliant MCP server. Anyone can paste your URL into Claude, Cursor or any OAuth-capable MCP client and start using it — you set the price per seat. ### What it does - **Standards-compliant** — Full MCP spec support including OAuth 2.1 sign-in, so modern clients connect without custom setup. - **Try before you buy** — Your server's tools are listable without signing in — prospects paste the URL, see what it does, and OAuth kicks in only when they call a tool. - **Knowledge as a tool** — Your knowledge bases are exposed as a search tool automatically — clients get grounded answers from your content. - **Metered and billed** — Per-seat subscriptions with Stripe billing built in. Usage is metered per call, with cost caps to protect you. ### How it works - **Assemble** — Pick the tools and knowledge bases your server exposes — build them once, reuse them across products. - **Price it** — Free, or per-seat monthly — Stripe checkout and invoicing are handled for you. - **Share the URL** — One endpoint works in every MCP client. Buyers connect in under a minute. ## Knowledge https://supercognit.com/product/knowledge Your content, answering questions Upload documents, link Google Docs, or point us at your website. Your agents and MCP servers answer from that content with hybrid retrieval and inline citations — not from the model's imagination. ### What it does - **Every format that matters** — PDF, Word, Markdown and plain text uploads, Google Docs links, and full website crawling with automatic extraction. - **Hybrid retrieval** — Semantic search combined with keyword matching, sentence-aware chunking and relevance floors — tuned so wrong answers get filtered, not served. - **Citations by default** — Answers cite the exact source passage, so your users can check the receipts. - **Fresh automatically** — Linked URLs re-sync on a schedule you choose — hourly on higher plans — so the agent never quotes last quarter's pricing. - **Reusable libraries** — Knowledge bases are workspace-level: build once, attach to any number of agents and MCP servers. ### How it works - **Ingest** — Drag in files, paste Google Docs links, or crawl your site. - **Attach** — Connect knowledge bases to any agent or MCP server in your workspace. - **Trust the answers** — Every response is grounded and cited; re-ingestion keeps sources current. ## Tools & automation https://supercognit.com/product/tools Agents that do things, not just say things Give your agent real capabilities: call any HTTP API, drive a real browser for the sites that don't have one, search the web, and take payments — each tool metered and permission-scoped. ### What it does - **HTTP tool builder** — Wrap any REST API as an agent tool with templated inputs — no code, SSRF-guarded by design. - **Browser automation** — A managed cloud browser completes multi-step website tasks — form fills, lookups, bookings — and reports back with results and PDFs. - **Web search** — Agents can search the live web when your knowledge doesn't cover the question, with sources attached. - **Payments** — Agents can request payment mid-conversation — checkout links generated and reconciled through the platform's billing. - **Tool library** — Tools are workspace-level and reusable across every agent and MCP server you publish. ### How it works - **Build or pick** — Create an HTTP tool from any API, or enable built-ins like browser tasks and web search. - **Scope it** — Each tool is permission-scoped and metered per call, with cost caps you control. - **Let the agent work** — The agent plans multi-step tool use on its own and reports what it did. ## Embeddable widget https://supercognit.com/product/widget One script tag, instant chat on your site Add one line of HTML and your agent lives on your website. The widget preloads silently in the background, so the first open is instant — no spinner, no iframe flash. ### What it does - **One-line install** — A single script tag with your agent's slug. Works on any site — WordPress, Shopify, hand-built, anything. - **Instant first open** — The chat iframe preloads hidden while the page renders, so opening the widget feels native. - **Programmable** — A small JavaScript API (open, close, toggle, events) lets your own buttons and flows drive the chat. - **Honest analytics** — Usage counts real opens — preloaded-but-never-opened embeds don't inflate your numbers. ### How it works - **Copy the snippet** — Grab your embed code from the dashboard — it's one script tag. - **Paste it** — Add it before on any page. The bubble appears, preloaded. - **Style and trigger** — Configure teaser messages and colors, or open it from your own UI via the API. ## Custom domains https://supercognit.com/product/domains chat.yourcompany.com, not ours Serve your agent from your own domain. Point DNS once — certificates, edge routing and tenant resolution are provisioned automatically, and your brand stays front and center. ### What it does - **Automatic certificates** — TLS is issued and renewed for you via DNS delegation — no certificate files, no expiry surprises. - **Edge-routed** — Traffic rides a global edge network to our origin; your visitors get the nearest point of presence. - **White-label surface** — The chat page carries your name, your logo, your domain — SuperCognit stays invisible to your users. ### How it works - **Connect** — Enter your domain in the dashboard and add two DNS records. - **Verify** — Certificates and routing provision automatically — live in minutes, not tickets. - **Ship it** — Your agent answers on your URL, indistinguishable from software you built in-house. ## WhatsApp & Telegram https://supercognit.com/product/channels Your agent, on WhatsApp and Telegram Connect a channel and the agent you already built starts answering there - same prompt, same knowledge, same tools, same inbox. No second bot to maintain and no separate script to keep in sync. ### What it does - **One agent, several doors** — A channel is a surface, not a copy. Edit the prompt once and every channel changes with it - there is no drift between what your website says and what WhatsApp says. - **Your number, your history** — You connect your own WhatsApp Business account and Telegram bot, so the number your customers know and the conversations they had remain yours. - **Documents welcome** — Customers send the PDF instead of describing it. The agent reads the attachment, answers from it, and keeps it with the conversation. - **Nothing falls between channels** — Every message becomes a conversation and a contact in the same inbox, whichever app it arrived from. ### How it works - **Connect** — Paste your WhatsApp Cloud API or Telegram credentials on the agent's Channels tab. - **Verify** — We register the webhook and confirm the round trip, so you see it working before a customer does. - **Answer** — Messages start arriving in the inbox, and the agent replies in the sender's language. ## CRM & lead capture https://supercognit.com/product/crm From conversation to CRM contact Connect Pipedrive or HubSpot and the agent stops being a chat log. Every qualified conversation becomes a contact your team can act on, with the exchange that produced it attached. ### What it does - **Qualification before handoff** — The agent gathers what your team needs to act, in the flow of the conversation, instead of dropping a raw transcript on a rep to triage. - **We are not your CRM** — No competing pipeline to keep in sync. We write contacts and notes into the system your team already lives in, and stay out of the way. - **Connected in clicks** — OAuth from your own CRM account. Connect it, pick the agent, done - and revoke it from the same place whenever you want. - **Never only in one place** — Leads are kept in your dashboard as well, so a CRM outage or a revoked token costs you nothing. ### How it works - **Connect** — Authorise Pipedrive or HubSpot from Integrations - OAuth, with no API keys to paste. - **Say what to ask** — Set what has to be collected before a conversation counts as a lead. - **Work the contacts** — Qualified conversations appear in your CRM, transcript attached, as they happen. ## Inbox https://supercognit.com/product/inbox Every conversation, and a team ready to take over An agent that answers in public and reports nothing back is a black box. The inbox is the other half: what was asked, what was answered, who asked it - and, when the agent cannot help, a person from your team stepping into the same conversation. ### What it does - **One list, every surface** — Website, widget, custom domain, WhatsApp and Telegram conversations arrive in the same place, against the same contacts. - **Take over in the same chat** — Reply from the inbox and the answer reaches the visitor on the web, WhatsApp or Telegram. The agent waits while a person is on it. - **Hand-over on your terms** — A "Talk to a person" button, hand-over when the agent is stuck, business hours, and a return to the AI if nobody answers. - **Work it as a team** — Assign, prioritise, tag and resolve, leave internal notes, and answer the usual questions with saved replies. - **Never miss a waiting visitor** — New hand-overs go to whoever is online, with email alerts, an Inbox badge and desktop notifications. - **The visitor, not just the text** — Country, language, referring page and returning-visitor history sit beside the transcript, so you can tell a browser from a buyer. ### How it works - **Turn on human support** — In the agent's Support tab, choose when it hands over and set your business hours. - **Invite your team** — Give support staff the Agent role: the inbox, and nothing else. - **Take over** — Reply when a visitor needs a person, then resolve it or let the agent carry on. ## Contacts https://supercognit.com/product/contacts The people behind the conversations A transcript tells you what was said. A contact tells you who said it, what they asked last time, and how to reach them - built automatically from every conversation, on every channel. ### What it does - **Built without asking** — Every conversation creates or updates a record. Nobody fills in a form and nobody imports a spreadsheet. - **Returning visitors resolve** — The same person across sessions, devices and channels lands on one record instead of scattering into anonymous fragments. - **Context for whoever replies** — Language, country, channel and prior questions are on the record, so a human follow-up is informed from the first line. - **Portable by default** — Export them, or sync them to your CRM. They are your customer relationships, not ours. ### How it works - **Publish an agent** — Contacts start being created from the first conversation, with no setup. - **Watch them fill in** — Each return visit and each channel adds to the record rather than creating a new one. - **Act on them** — Follow up from the dashboard, export the list, or sync it into your CRM. ## Lead capture https://supercognit.com/product/leads Leads, not chat logs A chat that ends with a satisfied visitor and no record is a missed sale. Set what qualifies a prospect and the agent gathers it in conversation, then hands you a lead with the exchange that produced it. ### What it does - **Answer first, ask second** — The agent earns the details by being useful, instead of gating help behind a form nobody wants to fill in. - **You set the bar** — Define what a lead has to include for your business. Below the bar it stays a contact; above it, it becomes something your team works. - **Evidence attached** — Every lead arrives with the conversation that produced it, so nobody has to ask the customer to explain twice. - **Wherever they came from** — The website, the widget, a custom domain, WhatsApp or Telegram - qualification works the same on all of them. ### How it works - **Define a lead** — Say what the agent must collect before a conversation counts as one. - **Let it work** — The agent answers questions and gathers details in the flow of the conversation. - **Route them** — Work them in the dashboard, or push them to your CRM or your own system. ## Analytics https://supercognit.com/product/analytics What your agents are actually doing Publishing an agent is the start, not the finish. Analytics tell you how much work it is absorbing, which questions defeat it, and what each answer costs - the three numbers that decide what you change next. ### What it does - **Demand, in your customers' words** — The questions people actually type are better product research than any survey, and they arrive for free. - **Gaps ranked by frequency** — Unanswered questions are sorted by how often they come up, so the next page you write is the one that pays the most. - **Per agent, not just per account** — Break volume, outcomes and cost down by agent to see which ones deserve more of your attention. - **The same meter that bills you** — Usage here is the usage you pay for - one number, not an estimate that drifts from the invoice. ### How it works - **Publish** — Metrics start collecting from the first conversation, with nothing to instrument. - **Read the gaps** — Work down the unanswered questions and add what is missing to your knowledge. - **Watch it improve** — The same questions stop appearing, and the resolved share climbs. ## Webhooks https://supercognit.com/product/webhooks Events, pushed to you When a conversation ends, a lead qualifies or a browser task finishes, your systems should not have to find out by polling. Point a webhook at your endpoint and the event arrives as it happens. ### What it does - **The escape hatch from our integrations** — You are never limited to the CRMs we happen to support. Anything that accepts an HTTP request can receive the same data. - **Documented payloads** — Every event type ships with its shape and a sample, so you build against something concrete rather than reverse-engineering a first delivery. - **One endpoint per job** — Subscribe each endpoint to only the events it needs, instead of filtering a firehose on your side. - **Pull as well as push** — The REST API covers what webhooks do not, so you can reconcile state as well as react to events. ### How it works - **Add an endpoint** — Point it at your URL in Integrations and choose the events it should receive. - **Check the sample** — Read the documented payload for each event and build your handler against it. - **Go live** — Events start arriving as conversations, leads and tasks happen. ## Workspaces & roles https://supercognit.com/product/workspaces Built for teams and agencies Everything lives in a workspace: agents, knowledge, tools, contacts, conversations, API keys and billing. Invite your team into it with roles, or keep one workspace per client with billing to match. ### What it does - **Clean client separation** — One workspace per client keeps knowledge, conversations and invoices apart, with no risk of the wrong agent seeing the wrong documents. - **Roles, not shared passwords** — Invite by email and set what each member can do - and take it away in one click when they move on. - **The workspace owns the work** — Agents and transcripts belong to the workspace, not to whoever created them, so staff changes cost you nothing. - **Billing that matches the boundary** — Plans and usage balances are per workspace, which is what makes per-client invoicing straightforward. ### How it works - **Create a workspace** — One per client, per team or per project - your personal one already exists. - **Invite the team** — Add people by email with the role they should have. - **Build inside it** — Agents, knowledge, contacts and billing all stay scoped to that workspace. ## Multilingual https://supercognit.com/product/languages One agent, every language Your customers do not all write in the same language, and maintaining one agent per language is how those agents start contradicting each other. Write it once: the interface ships in 12 languages and replies follow the visitor. ### What it does - **One prompt, many languages** — No per-language duplicates to keep in sync, and no chance of the Portuguese version promising something the English one does not. - **Follows the visitor** — Whatever language the message arrives in is the language the answer goes back in - nobody has to find a flag to click. - **Knowledge stays as written** — Documents in one language, answers in another. Your content does not need a translation project first. - **Exact where it matters** — Per-agent translations pin the wording you cannot leave to chance, while everything else adapts on its own. ### How it works - **Write once** — One prompt and one knowledge base, in whatever language suits you. - **Publish** — The chat interface picks up the visitor's language automatically. - **Pin what matters** — Add per-agent translations for the lines that must read exactly as you wrote them. # Developer API https://supercognit.com/api The AI API behind every SuperCognit agent One REST API to create agents and MCP servers, attach tools and knowledge, read every conversation and push events into your own systems. OpenAPI 3.1, key authentication, no SDK required. Base URL: https://api.supercognit.com/v1 Reference: https://api.supercognit.com/docs OpenAPI document: https://api.supercognit.com/v1/openapi.json ### Authenticate with a workspace key Create a key in the dashboard and send it as an X-Api-Key header. Keys carry read or write scope, and a key can be locked to a single product — so an integration only ever touches what you gave it. ### What the API covers - **Products** (`/v1/products`) — Create, update, publish and unpublish chat agents and MCP servers — prompt, summary, slug and price included. - **Tools** (`/v1/tools`) — Manage the workspace tool library: LLM, HTTP and static tools, their inputs, and which products use them. - **Knowledge bases** (`/v1/knowledge-bases`) — List the knowledge behind your answers and see which products each base is attached to. - **Conversations** (`/v1/sessions`) — Read conversations and full transcripts — every message, model and token count, ready for your analytics or your CRM. - **Webhooks** (`/v1/webhooks`) — Register endpoints, inspect deliveries and retries, and re-enable an endpoint without touching the dashboard. - **Wallet** (`/v1/wallet`) — Read the usage-credit balance and every top-up, charge and refund behind it. - **End users** (`/v1/verify-user`) — Verify a chat session token server-side, so your own app can trust who it is talking to. ### What teams build with it - **An agent per customer** — Agencies and SaaS teams create, configure and publish one agent per client straight from their own onboarding flow — no clicking through the dashboard. - **Conversations in your CRM** — Stream every conversation and captured contact into HubSpot, Salesforce or your own database as it happens. - **Knowledge that stays current** — Refresh a knowledge source when your docs change, so answers follow your product instead of trailing behind it. - **Usage you can bill on** — Read the wallet and the transcripts to see exactly what an account consumed, then charge for it inside your own product. ### Events, pushed to you Register an endpoint and SuperCognit POSTs JSON as conversations happen. Every request is signed with HMAC-SHA256 over the raw body, retried up to five times with exponential backoff, and kept in a delivery log with its response status. - conversation.created — someone started a new conversation - message.created — a message was sent, by a visitor or by the agent - message.completed — the agent finished its answer ### REST for your systems, MCP for AI clients The API is how your code drives SuperCognit. MCP is how AI clients — Claude, Cursor, your own agents — use what you publish. Both run on the same workspace, so a tool you build once is reachable either way. ### Questions developers ask **Is the API included in my plan?** Yes. Every plan includes API access at no extra charge — you pay your plan and the usage your agents generate, exactly as you would from the dashboard. The 7-day free trial includes it too. **How do I authenticate?** With a workspace API key, sent as the X-Api-Key header. Keys declare scopes — write is required for anything that is not a GET — and a key can be scoped to a single product, in which case it cannot touch workspace-level resources such as tools, knowledge bases or webhooks. **Do I need an SDK?** No. It is plain REST over JSON, so any HTTP client works. The OpenAPI 3.1 document is public, so you can also generate a typed client in the language of your choice in a couple of minutes. **Can I create and publish an agent entirely through the API?** Yes. Create the product, attach tools and knowledge bases, then publish it — the same lifecycle the dashboard uses. That is how teams provision one agent per customer automatically. **Which AI model does it use?** The platform runs a curated model behind your agents, so your integration never handles provider keys or model names. You build and ship the agent; we keep the model current underneath it. **How do I get conversations out of SuperCognit?** Two ways: read the conversation endpoints for sessions and transcripts, or register a webhook and receive each message as it happens. Most teams use webhooks for realtime and the endpoints for backfill. # Comparisons ## SuperCognit vs Chatbase https://supercognit.com/compare/chatbase Chatbase is a popular no-code builder for chatbots trained on your data, aimed mostly at website support. Chatbase made 'train a chatbot on your data' mainstream, and it does that well. SuperCognit starts where a support chatbot ends: agents that use tools and take actions, publish to their own domain, answer on WhatsApp and Telegram — and that you can sell as a product with billing built in. | Capability | SuperCognit | Chatbase | | --- | --- | --- | | No-code agent builder | Yes — describe it, publish it | Yes | | Train on your website & documents | Yes, with citations in answers | Yes | | Multi-step tool use (real actions) | Yes — HTTP tools, web search, browser automation | Limited — mostly Q&A and forms | | Sell your agent (checkout, seats, payouts) | Built in — set a price, get paid | Not a platform focus | | MCP server per workspace | Yes — Claude and Cursor can call your tools | No | | Custom domain with SSL | Yes — your agent on your domain | Embed-focused | | WhatsApp & Telegram channels | Yes, same agent everywhere | Integrations available | | End-user languages | 12 languages out of the box | Multilingual replies | | Cost controls | Per-seat and global caps, metered calls | Message credits per plan | | Free trial | 7 days, full platform | Free tier with limits | ### Why teams choose SuperCognit - **Agents that do things, not just answer** — SuperCognit agents run a genuine multi-step tool loop: they can call your APIs, search the web, drive a browser and check knowledge — in one conversation turn. That's the difference between deflecting a ticket and completing a task. - **A product, not a widget** — Your agent gets its own domain, its own brand, its own price. Chatbase ends at the embed; SuperCognit ends at a paying subscriber. - **MCP-native from day one** — Every SuperCognit workspace can expose its tools as an MCP server, so your product plugs into Claude, Cursor and the growing MCP ecosystem — a distribution channel chat widgets simply don't have. **Verdict:** If you need a support Q&A bubble on one website, Chatbase is a fine choice. If you want an AI product — on your domain, on messaging channels, with tools and revenue — SuperCognit is built for exactly that. ## SuperCognit vs Botpress https://supercognit.com/compare/botpress Botpress is a developer-oriented platform for building conversational AI with a visual flow editor. Botpress is powerful in the hands of a developer: flows, hooks, and code when you need it. SuperCognit takes a different bet — that the person with the knowledge shouldn't need a developer at all. You write what the agent does in plain language; knowledge, tools, channels, domains and billing are switches, not projects. | Capability | SuperCognit | Botpress | | --- | --- | --- | | Setup | Prompt-first, minutes to publish | Flow builder, steeper learning curve | | Who it's for | Experts, agencies, teams — no code | Developers and technical teams | | Knowledge with citations | Built in — files, Google Docs, website import | Knowledge bases available | | Sell your agent (checkout, seats, payouts) | Built in | Bring your own billing | | MCP server per workspace | Yes | Not a core feature | | Custom domain with SSL | Yes, automated | Self-managed hosting options | | WhatsApp & Telegram channels | Yes, native | Yes, via integrations | | Cost controls | Per-seat and global caps, metered calls | Usage-based, self-managed | | Browser automation tool | Yes — agents can drive real websites | Custom code required | | Free trial | 7 days, full platform | Free tier for builders | ### Why teams choose SuperCognit - **No flows to maintain** — Flow charts age badly: every new intent is another branch. SuperCognit agents reason over instructions, knowledge and tools, so behavior changes are a prompt edit — not a diagram refactor. - **Business model included** — Botpress gives you a bot; monetizing it is your problem. On SuperCognit, pricing, checkout, seats and payouts are part of publishing. - **Faster path for non-developers** — Agencies use SuperCognit to ship client agents in an afternoon — website import builds the knowledge base, the wizard drafts the agent, and the widget or domain goes live the same day. **Verdict:** Choose Botpress if you have engineers who want full control of conversation flows. Choose SuperCognit if you want the outcome — a published, monetizable agent with tools and knowledge — without the engineering project. ## SuperCognit vs Voiceflow https://supercognit.com/compare/voiceflow Voiceflow is a conversation-design platform used by teams to prototype and ship chat and voice assistants. Voiceflow shines at conversation design: canvases, collaboration, and control over every turn. SuperCognit optimizes for a different goal — going from knowledge to a live, revenue-ready agent fast, and letting the model handle the conversation instead of a hand-drawn canvas. | Capability | SuperCognit | Voiceflow | | --- | --- | --- | | Setup | Prompt-first, minutes to publish | Design canvas, per-turn control | | Voice assistants | Chat-first (web, WhatsApp, Telegram) | Yes — voice is a strength | | Knowledge with citations | Built in — files, Google Docs, website import | Knowledge base available | | Sell your agent (checkout, seats, payouts) | Built in | Bring your own billing | | MCP server per workspace | Yes | No | | Custom domain with SSL | Yes, automated | Embed/API focus | | Multi-step tool use | HTTP tools, web search, browser automation | API steps in flows | | End-user languages | 12 languages out of the box | Multilingual support | | Team collaboration | Workspaces with roles | Yes — strong design collaboration | | Free trial | 7 days, full platform | Free tier with limits | ### Why teams choose SuperCognit - **The model is the designer** — Modern models handle dialogue better than any canvas. You define goals, guardrails, knowledge and tools; the agent handles the ten thousand conversation paths you'd otherwise have to draw. - **From design tool to distribution** — SuperCognit publishes to your domain, an embeddable widget, WhatsApp, Telegram and MCP — with pricing attached. It's the whole pipeline from idea to paying users, not just the design phase. - **Built to be sold** — If your assistant is a product — for clients or customers — checkout, seats and payouts are already wired in. **Verdict:** Voiceflow is a great fit for teams designing voice experiences turn by turn. If your goal is a published, monetized, tool-using chat agent this week, SuperCognit gets you there with far less assembly. ## SuperCognit vs CustomGPT.ai https://supercognit.com/compare/customgpt CustomGPT.ai builds RAG chatbots from your content, with a focus on accurate, cited answers. CustomGPT.ai is a solid 'answers from your content' engine. SuperCognit includes that — website import, documents, citations — and keeps going: tools that take actions, messaging channels, your own domain, an MCP server, and native monetization so the agent can be a business, not just a search box. | Capability | SuperCognit | CustomGPT.ai | | --- | --- | --- | | RAG on your content with citations | Yes — hybrid retrieval, citations | Yes — core strength | | Website import | Yes — crawl, build KB, draft the agent | Yes | | Multi-step tool use (real actions) | Yes — HTTP tools, web search, browser automation | Limited — answer-focused | | Sell your agent (checkout, seats, payouts) | Built in | Not a platform focus | | MCP server per workspace | Yes | No | | Custom domain with SSL | Yes, automated | Embed/share links | | WhatsApp & Telegram channels | Yes, native | Limited | | End-user languages | 12 languages out of the box | Multilingual answers | | Cost controls | Per-seat and global caps, metered calls | Query credits per plan | | Free trial | 7 days, full platform | Trial available | ### Why teams choose SuperCognit - **Answers are the floor, not the ceiling** — Accurate cited answers are table stakes on SuperCognit — the same agent can also call your booking API, search the live web, or drive a browser to complete a task for the user. - **Everywhere your users are** — A knowledge bot that only lives in an embed misses the channels people actually use. SuperCognit agents answer on your site, on WhatsApp and on Telegram — the same brain everywhere. - **Monetize the expertise** — If your content is worth asking questions about, it's worth charging for. Set a price and SuperCognit handles subscriptions, seats and payouts. **Verdict:** For an internal 'ask our docs' bot, CustomGPT.ai does the job. For a customer-facing product with actions, channels, your domain and revenue, SuperCognit is the more complete platform. ## SuperCognit vs Dante AI https://supercognit.com/compare/dante-ai Dante AI offers no-code AI chatbots and avatars trained on your data. Dante AI makes it easy to stand up a chatbot on your data. SuperCognit aims at the step after that: an agent that acts — with tools, live knowledge, messaging channels and its own domain — and that you can turn into recurring revenue without any billing code. | Capability | SuperCognit | Dante AI | | --- | --- | --- | | No-code setup | Yes — prompt-first wizard | Yes | | Train on files & websites | Yes, with citations and auto-refresh | Yes | | Multi-step tool use (real actions) | Yes — HTTP tools, web search, browser automation | Limited | | Sell your agent (checkout, seats, payouts) | Built in | Not a platform focus | | MCP server per workspace | Yes | No | | Custom domain with SSL | Yes, automated | Embed/share links | | WhatsApp & Telegram channels | Yes, native | Integrations available | | AI avatars | No — chat-first | Yes | | Cost controls | Per-seat and global caps, metered calls | Message credits per plan | | Free trial | 7 days, full platform | Free tier with limits | ### Why teams choose SuperCognit - **Substance over surface** — Avatars are a nice demo; retention comes from an agent that resolves real requests. SuperCognit invests in the loop underneath — tools, retrieval quality, cost controls — so the agent keeps being useful after week one. - **Knowledge that stays fresh** — Imported URLs re-sync on a schedule, so your agent answers from today's website, not last quarter's crawl. - **A revenue line, not a line item** — Publish with a monthly price and SuperCognit runs checkout, seats, invoices and payouts — your expertise becomes a subscription product. **Verdict:** If a friendly avatar chatbot covers your need, Dante AI can do it. If you're building something users rely on — and pay for — SuperCognit gives you the tools, channels and billing to make it real. ## SuperCognit vs Tidio https://supercognit.com/compare/tidio Tidio is a customer-service suite with live chat and Lyro, its AI support agent. Tidio is a helpdesk with AI attached: shared inbox, live chat, and Lyro answering common questions. SuperCognit approaches from the opposite direction — the AI agent is the product. You build it on your knowledge and tools, publish it on your domain and channels, and optionally charge for access. There's an inbox here too; it just isn't the point. | Capability | SuperCognit | Tidio | | --- | --- | --- | | Primary focus | Build & monetize AI agents | Customer-service suite | | Human live-chat handoff | Yes — your team replies from the inbox while the AI waits | Yes — core strength | | Train on your website & documents | Yes, with citations | Yes, for Lyro | | Multi-step tool use (real actions) | Yes — HTTP tools, web search, browser automation | Support-task focused | | Sell your agent (checkout, seats, payouts) | Built in | No | | MCP server per workspace | Yes | No | | Custom domain with SSL | Yes — the agent is a site | Widget on your site | | WhatsApp & Telegram channels | Yes, native | Yes, plus Messenger/Instagram | | Pricing model | Plans + metered usage with caps | Per-seat + per-conversation | | Free trial | 7 days, full platform | Free plan with limits | ### Why teams choose SuperCognit - **An agent, not a deflection layer** — Lyro's job is to reduce tickets. A SuperCognit agent's job is whatever you give it: concierge, advisor, qualifier, operator — with tools to actually complete tasks, not just answer about them. - **Your brand, your domain, your price** — SuperCognit agents stand on their own: custom domain, your logo and colors, optional subscription pricing. It's a product surface, not a support widget. - **Built for builders and agencies** — Workspaces, wholesale partner pricing and per-client products make SuperCognit the platform you build a portfolio on — something a helpdesk suite was never designed for. **Verdict:** If you're staffing a support team and want AI to deflect tickets, Tidio fits. If you want to build the AI itself — and possibly sell it — that's what SuperCognit is for. ## SuperCognit vs Intercom Fin https://supercognit.com/compare/intercom-fin Intercom Fin is the AI support agent inside Intercom's customer-service platform, priced per resolved conversation. Fin is a strong support agent — if you already live inside Intercom. It answers from your help center and charges per resolution, on top of the Intercom seats underneath. SuperCognit starts from a different place: the agent is the product, it lives on your own domain and channels, it takes actions beyond support deflection, and you can sell access to it. | Capability | SuperCognit | Intercom Fin | | --- | --- | --- | | Primary focus | Build & monetize AI agents | Support automation inside Intercom | | Pricing shape | Plans + metered usage with caps | Per resolution, plus Intercom seats | | Works without a helpdesk suite | Yes — standalone platform | Designed for Intercom | | Multi-step tool use (real actions) | Yes — HTTP tools, web search, browser automation | Support-task focused | | Sell your agent (checkout, seats, payouts) | Built in | No | | MCP server per workspace | Yes | No | | Custom domain with SSL | Yes — the agent is a site | Messenger widget on your site | | WhatsApp & Telegram channels | Yes, native | WhatsApp via Intercom | | Human handoff & inbox | Yes — your team replies from the inbox while the AI waits | Yes — core strength | | Free trial | 7 days, full platform | Trial via Intercom plans | ### Why teams choose SuperCognit - **No per-resolution meter** — Per-resolution pricing means your bill grows exactly when your agent gets good. SuperCognit plans meter usage with hard cost caps — success doesn't invoice you per answered question. - **Bigger job than deflection** — Fin closes tickets. A SuperCognit agent can be a concierge, advisor, qualifier or operator — calling your APIs, searching the web and driving a browser, not just resolving support threads. - **Yours to brand and sell** — Your agent stands on your own domain with your brand, plugs into Claude and Cursor over MCP, and can carry a subscription price. Fin is a feature of Intercom; your SuperCognit agent is a product of yours. **Verdict:** If your team already runs on Intercom and support deflection is the whole goal, Fin is a natural fit. If you want an agent that's a standalone, monetizable product on your own domain and channels, SuperCognit is built for that. ## SuperCognit vs Chatling https://supercognit.com/compare/chatling Chatling is a budget-friendly no-code builder for AI chatbots trained on your content. Chatling makes it cheap and easy to get a chatbot answering from your content. SuperCognit covers that ground — website import, cited answers, a widget — and then keeps going where lightweight builders stop: real tool use, WhatsApp and Telegram, your own domain, an MCP server, and native monetization. | Capability | SuperCognit | Chatling | | --- | --- | --- | | No-code setup | Yes — prompt-first wizard | Yes — quick and simple | | Train on your website & documents | Yes, with citations and auto-refresh | Yes | | Multi-step tool use (real actions) | Yes — HTTP tools, web search, browser automation | Limited — Q&A focused | | Sell your agent (checkout, seats, payouts) | Built in | No | | MCP server per workspace | Yes | No | | Custom domain with SSL | Yes, automated | Embed/share links | | WhatsApp & Telegram channels | Yes, native | Limited | | End-user languages | 12 languages out of the box | Multilingual replies | | Cost controls | Per-seat and global caps, metered calls | Message credits per plan | | Free trial | 7 days, full platform | Free tier with limits | ### Why teams choose SuperCognit - **Room to grow into** — A cheap Q&A widget is fine until the day you need the agent to check an order, push a lead to your CRM or answer on WhatsApp. On SuperCognit those are switches, not a platform migration. - **A channel strategy, not just an embed** — The same agent answers on your site, on your own domain, on Telegram and WhatsApp, and over MCP to Claude and Cursor — one brain, every door. - **Turn the bot into a business** — If what your bot knows is valuable, charge for it: pricing, checkout, seats and payouts are built into publishing. **Verdict:** For a simple, low-cost website Q&A bot, Chatling does the job. If the bot is meant to become a real product — with actions, channels and revenue — SuperCognit gives it somewhere to go. ## SuperCognit vs Wonderchat https://supercognit.com/compare/wonderchat Wonderchat builds no-code AI chatbots from your website and docs, embedded on your site in minutes. Wonderchat nails the quick start: point it at your site, get a chatbot. SuperCognit's importer gives you the same minutes-to-live start — and then hands you the rest of the platform: tools that act, messaging channels, your own branded domain, an MCP server, and billing if you want to charge for access. | Capability | SuperCognit | Wonderchat | | --- | --- | --- | | Website import → live bot | Yes — minutes, with a drafted agent | Yes — minutes | | Answers with citations | Yes — hybrid retrieval | Yes | | Multi-step tool use (real actions) | Yes — HTTP tools, web search, browser automation | Limited — answer focused | | Sell your agent (checkout, seats, payouts) | Built in | No | | MCP server per workspace | Yes | No | | Custom domain with SSL | Yes, automated | Embed/share links | | WhatsApp & Telegram channels | Yes, native | Limited | | Human handoff & inbox | Yes — team takeover in the inbox, plus CRM | Basic handoff | | Cost controls | Per-seat and global caps, metered calls | Message quotas per plan | | Free trial | 7 days, full platform | Free trial available | ### Why teams choose SuperCognit - **Same fast start, longer runway** — Both platforms get a bot live from your website in minutes. On SuperCognit that first bot can grow into an omnichannel agent with tools, a domain and a price — without starting over elsewhere. - **Answers that lead somewhere** — Citations are table stakes; SuperCognit agents also capture leads into the built-in CRM, call your APIs, and hand off to your team with full conversation history. - **Distribution beyond the bubble** — A chat bubble reaches visitors who find your site. Your own domain, WhatsApp, Telegram and MCP reach everyone else. **Verdict:** Wonderchat is a solid pick for a fast website chatbot. If you want that same fast start on a platform that scales into channels, actions and revenue, start on SuperCognit. ## SuperCognit vs Zendesk AI https://supercognit.com/compare/zendesk-ai Zendesk AI adds AI agents and copilots to Zendesk's enterprise customer-service suite. Zendesk AI is deep ticketing automation for teams committed to the Zendesk suite, with pricing built around resolutions and seats. SuperCognit isn't a helpdesk — it's the platform where the AI agent itself is the product: built from your knowledge, published on your domain and channels, monetizable, and priced on plans with hard cost caps. | Capability | SuperCognit | Zendesk AI | | --- | --- | --- | | Primary focus | Build & monetize AI agents | Enterprise support suite automation | | Pricing shape | Plans + metered usage with caps | Seats plus per-resolution AI pricing | | Works without a helpdesk suite | Yes — standalone platform | Designed for Zendesk | | Setup | Import your website, publish in minutes | Enterprise onboarding | | Multi-step tool use (real actions) | Yes — HTTP tools, web search, browser automation | Ticket workflows and integrations | | Sell your agent (checkout, seats, payouts) | Built in | No | | MCP server per workspace | Yes | Not a core feature | | Custom domain with SSL | Yes — the agent is a site | Widget/help-center on your site | | WhatsApp & Telegram channels | Yes, native | Yes, via Zendesk messaging | | Free trial | 7 days, full platform | Suite trial available | ### Why teams choose SuperCognit - **Right-sized for builders** — Zendesk AI assumes a support organization: seats, queues, SLAs. SuperCognit assumes a builder with knowledge worth sharing — one person can publish a serious agent in an afternoon. - **Costs you can cap** — Per-resolution AI pricing on top of per-seat suites compounds. SuperCognit plans meter usage with per-seat and global caps, so a viral week can't torch the budget. - **The agent is the asset** — On Zendesk, AI makes your support cheaper. On SuperCognit, the agent itself can earn — published on your domain, connected over MCP, sold on subscription. **Verdict:** Large support organizations already on Zendesk will get value from its AI. If you're building an AI product rather than staffing a helpdesk, SuperCognit is the platform shaped for you. ## SuperCognit vs LiveChatAI https://supercognit.com/compare/livechatai LiveChatAI turns your knowledge base into an AI support agent with live-chat handoff, aimed at ticket deflection. LiveChatAI does the support-deflection job well: train on your content, answer common questions, hand off to a human. SuperCognit includes that pattern — and treats it as the starting point. Your agent gets tools that act, WhatsApp and Telegram channels, its own branded domain, an MCP server, and native monetization, so it can be a product rather than a support widget. | Capability | SuperCognit | LiveChatAI | | --- | --- | --- | | Primary focus | Build & monetize AI agents | Support ticket deflection | | Train on your website & documents | Yes, with citations and auto-refresh | Yes | | Multi-step tool use (real actions) | Yes — HTTP tools, web search, browser automation | Support-task focused | | Human live-chat handoff | Yes — your team replies from the inbox while the AI waits | Yes — core strength | | Sell your agent (checkout, seats, payouts) | Built in | No | | MCP server per workspace | Yes | No | | Custom domain with SSL | Yes — the agent is a site | Widget/share links | | WhatsApp & Telegram channels | Yes, native | Limited | | Cost controls | Per-seat and global caps, metered calls | Message/resolution quotas per plan | | Free trial | 7 days, full platform | Free tier with limits | ### Why teams choose SuperCognit - **Deflection is a feature, not the product** — Answering repeat questions is the floor. SuperCognit agents also call your APIs, search the live web and drive a browser — resolving the task, not just the ticket. - **Channels beyond the bubble** — A support widget only meets customers who found your site. The same SuperCognit agent answers on your own domain, on WhatsApp and Telegram, and over MCP inside Claude and Cursor. - **Revenue, not just savings** — Support AI saves money; a published agent can earn it. Set a price and SuperCognit runs checkout, seats and payouts. **Verdict:** If ticket deflection with human handoff is the whole brief, LiveChatAI covers it. If you want the agent to be a monetizable product on channels you own, SuperCognit is the more complete platform. ## SuperCognit vs Botsonic https://supercognit.com/compare/botsonic Botsonic (by Writesonic) is a no-code builder for custom AI chatbots trained on your data. Botsonic makes standing up a chatbot on your content straightforward, with the Writesonic ecosystem around it. SuperCognit matches the no-code start — website import, cited answers, an embeddable widget — and extends it into what a chatbot becomes next: an agent with real tools, messaging channels, its own domain, an MCP server and built-in monetization. | Capability | SuperCognit | Botsonic | | --- | --- | --- | | No-code setup | Yes — prompt-first wizard | Yes | | Train on your website & documents | Yes, with citations and auto-refresh | Yes | | Multi-step tool use (real actions) | Yes — HTTP tools, web search, browser automation | Limited — Q&A and forms focused | | Sell your agent (checkout, seats, payouts) | Built in | No | | MCP server per workspace | Yes | No | | Custom domain with SSL | Yes, automated | Embed/share links | | WhatsApp & Telegram channels | Yes, native | Integrations available | | End-user languages | 12 languages out of the box | Multilingual replies | | Cost controls | Per-seat and global caps, metered calls | Word/message credits per plan | | Free trial | 7 days, full platform | Free tier with limits | ### Why teams choose SuperCognit - **Credits don't cap your ambition** — Credit-metered plans make every conversation a budgeting question. SuperCognit meters usage with hard per-seat and global caps you control — predictable at any traffic level. - **An agent, not an add-on** — Botsonic lives inside a writing-tools suite. On SuperCognit the agent is the product: your domain, your brand, your price, your MCP endpoint. - **Room for the second act** — When the chatbot needs to check an order, book a slot or push a lead to your CRM, SuperCognit's HTTP tools and browser automation are already there — no platform change. **Verdict:** Botsonic is a reasonable pick if you're already in the Writesonic ecosystem and need a simple site chatbot. For an agent you can grow, distribute and sell, SuperCognit is built for the longer arc. ## SuperCognit vs Chatfuel https://supercognit.com/compare/chatfuel Chatfuel is a chatbot platform for Messenger, Instagram and WhatsApp, focused on marketing automation flows. Chatfuel comes from the social-messaging world: broadcast flows, keyword triggers and marketing automation on Meta's channels. SuperCognit approaches conversation from the other side — a knowledge-grounded AI agent that genuinely answers and acts — and then meets you on messaging channels too, including WhatsApp and Telegram, plus the web, your own domain and MCP. | Capability | SuperCognit | Chatfuel | | --- | --- | --- | | Primary focus | Knowledge-grounded AI agents | Social messaging marketing flows | | Setup | Prompt-first, minutes to publish | Flow builder per channel | | Train on your website & documents | Yes, with citations | Limited — flow/keyword centric | | Multi-step tool use (real actions) | Yes — HTTP tools, web search, browser automation | Flow actions and integrations | | Messenger & Instagram | Not yet — web, WhatsApp, Telegram | Yes — core strength | | WhatsApp | Yes, native | Yes | | Sell your agent (checkout, seats, payouts) | Built in | No | | MCP server per workspace | Yes | No | | Custom domain with SSL | Yes — the agent is a site | Channel-bound | | Free trial | 7 days, full platform | Trial available | ### Why teams choose SuperCognit - **Answers, not just automations** — Keyword flows break the moment a customer asks something unscripted. A SuperCognit agent reasons over your actual knowledge with citations — every question is in scope. - **Own the primary channel** — Building only on Meta's channels rents your audience — and WhatsApp charges per business reply from October 2026. SuperCognit puts your best answers on your own site and domain first, with messaging as reach. - **One brain everywhere** — The same agent, knowledge and inbox across web, WhatsApp and Telegram — no per-channel flow rebuilding. **Verdict:** For Messenger/Instagram marketing broadcasts, Chatfuel is purpose-built. For an AI agent that answers from your knowledge, acts, and lives on channels you own, SuperCognit is the stronger foundation. ## SuperCognit vs MyAskAI https://supercognit.com/compare/myaskai MyAskAI is an AI support agent trained on your help docs, positioned as a lightweight add-on to tools like Intercom and Zendesk. MyAskAI's pitch is pragmatic: bolt an AI answerer onto the support stack you already run, cheaper than the suites' own AI. SuperCognit plays a different game — it's the standalone platform where the agent itself is the product: knowledge, tools, channels, domain and billing in one place, no helpdesk suite required. | Capability | SuperCognit | MyAskAI | | --- | --- | --- | | Primary focus | Build & monetize AI agents | Support deflection add-on | | Works without a helpdesk suite | Yes — standalone platform | Best alongside Intercom/Zendesk | | Train on your website & documents | Yes, with citations and auto-refresh | Yes — help docs focused | | Multi-step tool use (real actions) | Yes — HTTP tools, web search, browser automation | Limited — answer focused | | Sell your agent (checkout, seats, payouts) | Built in | No | | MCP server per workspace | Yes | No | | Custom domain with SSL | Yes — the agent is a site | Widget/share links | | WhatsApp & Telegram channels | Yes, native | Limited | | Cost controls | Per-seat and global caps, metered calls | Conversation quotas per plan | | Free trial | 7 days, full platform | Trial available | ### Why teams choose SuperCognit - **A platform, not a patch** — An add-on inherits the limits of the stack underneath. SuperCognit is the whole surface: agent, knowledge, tools, inbox, channels and billing — one workspace, no suite tax. - **Beyond the help center** — Docs Q&A is one use case. The same SuperCognit agent qualifies leads, books appointments through your APIs, and sells access on subscription. - **Distribution the add-ons can't reach** — Your own domain, an embeddable widget, WhatsApp, Telegram and an MCP endpoint for Claude and Cursor — surfaces a helpdesk add-on doesn't touch. **Verdict:** If you run Intercom or Zendesk and just want cheaper deflection, MyAskAI is a sensible bolt-on. If you're building the agent as its own product, SuperCognit gives you the entire platform. # Customer stories ## How Four Seasons Fairways gave every guest a 24/7 multilingual concierge https://supercognit.com/customers/four-seasons-fairways The luxury Quinta do Lago resort turned its website into an AI concierge that answers guest questions day and night, in the guest's own language — imported and live in minutes on SuperCognit. - **Industry:** Hospitality — luxury resort - **Location:** Quinta do Lago, Algarve, Portugal - **Use case:** Guest concierge & booking assistant ### Challenge Four Seasons Fairways is a luxury resort of 132 villas and apartments in Quinta do Lago, with two restaurants, a wellness centre, championship golf nearby and a kids club. Its guests arrive from all over Europe and beyond — and their questions don't keep office hours. Villa layouts, restaurant menus, spa treatments, golf tee times, transfers, late check-out: the answers all existed on the resort's website, but guests had to dig for them, and the front desk answered the same questions dozens of times a day — often in a second or third language. ### Solution The resort's website was imported into SuperCognit: the crawler read the site, built a cited knowledge base from it, and drafted the concierge agent — live within minutes, not weeks. The team then tuned the agent's tone to match the Fairways brand and reviewed conversations in the inbox to fill knowledge gaps. - Website import built the knowledge base directly from fourseasonsfairways.com, with scheduled re-syncs so seasonal menus and offers stay current. - The concierge answers in the guest's language — English, Portuguese, German, French and more — with the same warm, informed tone. - Answers cite their sources, so guests can jump to the exact page for villa details or restaurant menus. - The embeddable widget put the concierge on the resort's own site with a single script tag. ### Results The concierge now handles the questions that used to queue at the front desk — availability windows, what's in a Deluxe Villa, whether the pool is heated, how to book a spa treatment — instantly, around the clock, in the guest's own language. Staff time shifts to the moments that need a human touch, and every conversation is visible in the SuperCognit inbox, so the team keeps learning what guests ask for most — before they even arrive. > Guests get answers at midnight in their own language, and the front desk starts the day with fewer emails — a concierge that knows the resort because it learned it from the resort’s own website. > — SuperCognit Team, on the Fairways deployment ## How a negotiation expert turned frameworks into a 24/7 advisor on SuperCognit https://supercognit.com/customers/negotiation-strategy-ai Negotiation expertise usually lives in one-off sessions and expensive workshops. This creator packaged frameworks, scripts and scenario playbooks into an AI advisor that prepares users for salary talks, business deals and supplier discussions — any hour, in their language. - **Industry:** Professional training & coaching - **Location:** Published on SuperCognit - **Use case:** Paid expert advisor — negotiation strategy ### Challenge Most people walk into negotiations without a structured plan — relying on intuition, generic advice, and hoping for the best. The result is predictable: weak positioning, premature concessions and money left on the table. The expertise to fix that exists, but it's traditionally locked inside 1:1 coaching sessions and workshops that most people can't access at the moment they actually need them — the night before the salary conversation. ### Solution Negotiation Strategy and Execution AI packages that expertise as a product. Built on SuperCognit, the agent delivers structured preparation, actionable frameworks, concrete scripts and scenario-based guidance across the situations where negotiation actually happens: - Salary and compensation negotiation — positioning, timing and counter-offer scripts. - Business, contract, client and supplier negotiations — deal structure and concession strategy. - Handling objections and pressure, closing strategies and outcome optimization. - Everyday negotiation — from cost reductions to vendor discussions. The creator wrote the agent's method in plain language, attached the knowledge, and published — SuperCognit handles the hosting, the multilingual conversations, and the metering underneath. ### Results What used to require booking a session is now available at the moment of need: users rehearse the exact conversation they're about to have, get scripts calibrated to their scenario, and come back before each new negotiation. For the creator, expertise that once sold by the hour now works as a product — one build, every user, around the clock. > Advice-by-the-hour caps an expert's reach at their calendar. Packaged as an agent, the same frameworks prepare someone for a salary talk at midnight — and the expert's product is working while they sleep. > — SuperCognit Team, on the Negotiation Strategy AI deployment ## How a footwear industry insider productized the global sourcing playbook https://supercognit.com/customers/footwear-sourcing-ai Finding the right footwear manufacturer takes years of industry knowledge — regions, production models, MOQs, RFQs. This creator turned that know-how into a sourcing consultant AI that guides brands and designers through the global supply chain, as part of a whole footwear product line. - **Industry:** Manufacturing & sourcing consultancy - **Location:** Published on SuperCognit - **Use case:** Paid expert advisor — footwear sourcing ### Challenge For a startup footwear brand, the distance between a design and a production run is enormous: which country fits your product type and price segment, private label versus OEM versus custom development, which suppliers make soles versus uppers versus packaging, what MOQs to expect, how to structure an RFQ that factories take seriously. That knowledge lives in the heads of industry veterans — and consulting hours don't scale across time zones and a global client base. ### Solution Footwear Sourcing Consultant AI encodes the sourcing playbook as a structured advisor. Built on SuperCognit, it guides users by footwear type, production volume, price segment and geography: - Manufacturing-region recommendations matched to product type and price positioning. - Production-model guidance — private label, OEM and custom development. - Supplier classification across factories, component makers and material providers, from soles and insoles to leather and packaging. - Supplier evaluation — quality considerations, certifications, minimum order quantities. - RFQ preparation — request-for-quotation structure and technical documentation. And it isn't alone: the same creator publishes a whole footwear line on the platform — brand building, video direction and sourcing — each agent its own product, all from one workspace. ### Results Emerging brands and designers get sourcing guidance that used to require knowing someone in the industry, at any hour, in their language. For the creator, one body of expertise became a repeatable product line — the model SuperCognit's workspaces were built for. > Niche industry knowledge is the most defensible content there is. A sourcing veteran's mental map of the global supply chain, packaged as an agent, is a product no generic chatbot can imitate. > — SuperCognit Team, on the Footwear Sourcing AI deployment ## How an admissions coach made essay and interview prep available to every student https://supercognit.com/customers/college-admissions-coach-ai Admissions coaching is seasonal, intense and expensive — exactly when students need the most help. This creator published an essay and interview coach that gives students structured, credible application prep on demand. - **Industry:** Education & coaching - **Location:** Published on SuperCognit - **Use case:** Paid expert advisor — admissions coaching ### Challenge Every application season, the same crunch: students staring at blank personal statements, recycling generic essay advice, and walking into interviews unrehearsed. Quality admissions coaching solves this — but it's expensive, concentrated in a few markets, and the coach's calendar physically can't fit every student in the weeks before deadlines. ### Solution College Admissions Essay and Interview Coach AI turns the coaching method into a product students can use whenever they're actually writing — which is usually late at night. Built on SuperCognit, it works the way a good coach does: positioning first, then narrative, then polish. - Applicant positioning and personal-narrative development — finding the story before writing the essay. - Personal statements and supplemental essays — structure, authentic voice, evidence of impact. - Interview preparation — common questions, explaining motivations and goals, avoiding the classic mistakes. - Feedback that sharpens the student's own writing rather than replacing it. ### Results Students get structured, credible prep at a fraction of traditional coaching economics — and at the hour they're actually working. The coach's method now reaches beyond their calendar and their city, packaged as a product that carries their approach into every conversation. > Deadline-season demand is exactly when a human coach is least available. An agent built on the coach's own method flattens that curve — every student gets the structured version of the advice, on time. > — SuperCognit Team, on the Admissions Coach AI deployment # Industries ## Hospitality & travel https://supercognit.com/industries/hospitality Hotels, resorts and travel brands answer the same questions hundreds of times a week — at every hour, in every language. A SuperCognit agent learns your property from your website and documents, speaks your guest's language, and is on duty around the clock: on your site, on WhatsApp, wherever your guests are. ### Use cases - **Guest concierge** — Rooms and villas, restaurant menus, spa treatments, kids club, transfers — answered instantly with citations to your own pages, in 12 languages. A guest who asks to see the rooms gets the property's video, played right in the chat. - **Booking assistant** — Qualify dates, party size and preferences, then hand the guest to your booking engine or capture the enquiry for your reservations team. - **Pre-arrival & in-stay support** — Check-in times, directions, what's on this week, late check-out — the questions that flood reception, deflected to a concierge that never sleeps. - **Staff copilot** — An internal agent trained on SOPs, menus and policies, so every seasonal hire answers like your best team member. ### How SuperCognit maps onto Hospitality & travel - **Website import → knowledge** — Point SuperCognit at your website; it crawls your pages, builds a cited knowledge base and drafts the agent. Scheduled re-syncs keep seasonal menus and offers current. - **Widget + custom domain** — One script tag adds the concierge to your site; or give it its own branded page on concierge.yourhotel.com with SSL handled for you. - **WhatsApp & Telegram channels** — Guests already message you on WhatsApp. Connect the same agent so every channel answers with the same knowledge and tone. - **HTTP tools** — Connect your booking engine or PMS API so the agent can check availability or create enquiries — actions, not just answers. - **Inbox & CRM** — Every conversation lands in your inbox with guest metadata, so your team sees what guests ask before they arrive. ### FAQ **How long does it take to launch a hotel concierge?** Minutes for a first working version: import your website, review the drafted agent, and publish to the widget or a domain. Four Seasons Fairways went from import to a live agent in minutes. **Can the agent answer in my guests' languages?** Yes — agents answer in 12 languages out of the box, automatically matching the language the guest writes in. **Can it connect to my booking system?** Yes. With the HTTP tool builder the agent can call your booking or PMS API — check availability, create enquiries, or hand off to your booking page at the right moment. ## Real estate https://supercognit.com/industries/real-estate Real-estate enquiries arrive at all hours and most start with the same questions: price, area, condition, availability, financing. A SuperCognit agent trained on your listings and neighbourhood guides answers instantly, qualifies the lead, and books the viewing — while your human agents do what only humans can. ### Use cases - **Listing Q&A** — Square metres, energy rating, condo fees, renovation state — answered from your listing data with citations, on every property page. - **Lead qualification** — Budget, timeline, financing status and must-haves collected conversationally, then delivered to your CRM or inbox as a structured lead. - **Area & buying-process guide** — Neighbourhood guides, taxes, buying steps for foreign buyers — the advisory content that builds trust before the first call. - **Agency knowledge copilot** — An internal agent on your procedures, mandates and legal templates, so new agents ramp up in days. ### How SuperCognit maps onto Real estate - **Knowledge bases** — Import your website and listing pages; upload PDFs like floor plans and market reports. Auto-refresh keeps listings current as they change. - **Widget on listing pages** — The embeddable widget puts the agent on every property page, preloaded and instant — one script tag across the whole portal. - **HTTP tools → your CRM** — Push qualified leads into your CRM the moment the conversation produces name, contact and intent. - **WhatsApp channel** — Buyers abroad ask on WhatsApp at odd hours. The same agent answers there, in their language, and logs the conversation. - **Custom domains** — Launch a branded advisor at homes.youragency.com — a product surface that markets itself. ### FAQ **Can the agent stay up to date as listings change?** Yes — imported URLs re-sync on a schedule, so sold properties drop out and new listings appear without manual re-training. **How does lead capture work?** The agent collects contact details and qualification answers naturally in conversation; every conversation lands in your inbox with metadata, and HTTP tools can push structured leads straight into your CRM. **Does it work for international buyers?** Yes — the agent answers in 12 languages automatically, which is exactly where a human team can't scale. ## E-commerce https://supercognit.com/industries/ecommerce Most abandoned carts start with an unanswered question: sizing, shipping, compatibility, returns. A SuperCognit agent trained on your catalog and policies answers in seconds — on the product page, at 2am, in the shopper's language — and hands your team only the conversations that need a human. ### Use cases - **Pre-sale product advisor** — "Will this fit?", "Is it compatible with…?", "When will it arrive?" — answered from your product data before the shopper leaves the page. - **Order & policy support** — Shipping, returns, warranties and sizing policies answered with citations to your own pages, consistently every time. - **Guided selling** — The agent asks what the shopper needs and recommends from your catalog — a personal shopper on every page. - **Post-purchase care** — Setup guides, care instructions and troubleshooting from your manuals, deflecting tickets that don't need an agent. ### How SuperCognit maps onto E-commerce - **Website & catalog import** — Crawl your store and help centre into a cited knowledge base; scheduled re-syncs keep prices and products current. - **HTTP tools → order status** — Connect your store's API so the agent can look up an order and answer 'where is my package?' with real data. - **Widget everywhere** — One script tag adds the advisor to every page — preloaded, instant, and styled to your brand. - **WhatsApp & Telegram** — Meet shoppers on the channels they already use; the same agent, the same catalog knowledge. - **Cost caps** — Per-seat and global cost caps mean a traffic spike never becomes a bill spike. ### FAQ **Can the agent check live order status?** Yes — connect your store's API with the HTTP tool builder and the agent can look up orders and shipping status in conversation. **What happens when the agent can't help?** Conversations land in your inbox with full history, so your team picks up exactly where the agent left off. **Does it handle multiple languages for international stores?** Yes — 11 end-user languages out of the box, automatically matching the shopper. ## Healthcare & clinics https://supercognit.com/industries/healthcare Clinics and healthcare providers spend hours a day on questions that aren't clinical: opening hours, which specialist to see, what to bring, how billing works, directions. A SuperCognit agent trained on your website and patient guides answers these instantly and consistently — and knows to route anything clinical to your team. ### Use cases - **Patient information** — Services, specialists, insurance, preparation instructions and opening hours — answered from your own materials, with citations. - **Appointment guidance** — Help patients find the right service and hand them to your booking system or front desk with context captured. - **Multilingual access** — Serve patients in 12 languages — tourists, expats and international patients get the same clarity as locals. - **Internal protocol copilot** — An internal agent on procedures and admin protocols so staff find the right document in seconds. ### How SuperCognit maps onto Healthcare & clinics - **Knowledge with guardrails** — The agent answers from your approved content and is instructed to defer clinical questions to your professionals — you control the boundaries in plain language. - **Website import** — Your services, team and FAQ pages become a cited knowledge base in minutes, re-synced on a schedule. - **Widget + custom domain** — On your site via one script tag, or on a dedicated info domain — SSL and hosting handled. - **HTTP tools** — Connect scheduling or contact systems so the agent can create appointment requests with the details already collected. - **Inbox oversight** — Every conversation is reviewable, so your team audits what patients are asking and refines the knowledge base. ### FAQ **Can we control what the agent will and won't answer?** Yes — behavior instructions are plain language, and the standard pattern is: answer practical/administrative questions from approved content, defer anything clinical to your team. **Is patient conversation data visible to us?** Yes — conversations land in your workspace inbox for review, and you control the agent's data collection behavior. **How fast can a clinic go live?** Import your website, review the drafted agent, adjust the guardrails, publish — a first version typically takes an afternoon, not a project plan. ## Education & coaching https://supercognit.com/industries/education Educators, course creators and coaches have the same bottleneck: you. Every student question competes for your hours. A SuperCognit agent trained on your materials answers like your teaching assistant — 24/7, in the student's language — and because monetization is native, access to it can be part of what students pay for. ### Use cases - **Course tutor** — Students ask about any lesson and get answers grounded in your materials, with citations back to the source module. - **Admissions & programme FAQ** — Requirements, dates, fees and curriculum questions answered instantly for schools and academies. - **Paid expert companion** — Coaches and consultants publish an agent trained on their frameworks — a subscription product students and clients pay for monthly. - **Practice partner** — Quizzing, worked examples and explanations on demand, so students practise between sessions instead of waiting for them. ### How SuperCognit maps onto Education & coaching - **Knowledge from your materials** — Upload course PDFs, link Google Docs, import your site — the agent masters your curriculum, not the open internet. - **Monetization built in** — Set a monthly price; SuperCognit runs checkout, seats, invoices and payouts. Your knowledge becomes a subscription product. - **Custom domain** — tutor.youracademy.com — your tutor as a standalone branded product, SSL included. - **Telegram channel** — Students live in messaging apps; your tutor can answer there with the same knowledge. - **Memory** — The agent remembers returning students' context, so help picks up where it left off. ### FAQ **Can I charge students for access to my agent?** Yes — that's native. Set a monthly price and SuperCognit handles checkout, seats and payouts; you can also keep it free or open only to logged-in users. **Will the agent stay faithful to my teaching?** It answers from your uploaded materials with citations, and its instructions — tone, method, boundaries — are yours to write in plain language. **What formats can I upload?** PDF, Word, text and Markdown files, Google Docs links, and full website imports. ## Professional services https://supercognit.com/industries/professional-services In professional services the product is expertise — and most of it is spent answering the same first-meeting questions. A SuperCognit agent trained on your guides and FAQs handles the routine layer: it explains your services, answers process questions, qualifies enquiries, and routes real matters to the right person with context attached. ### Use cases - **Client intake & qualification** — The agent collects matter type, urgency and background conversationally, so your first human touch starts informed. - **Process & fees FAQ** — "How does the process work?", "What documents do I need?", "How do you bill?" — answered consistently from your approved content. - **Internal knowledge copilot** — Playbooks, precedents, checklists and SOPs searchable in plain language — your associates' fastest colleague. - **Productized advice** — Package a niche expertise — expat tax, GDPR readiness, licensing — as a paid agent clients subscribe to. ### How SuperCognit maps onto Professional services - **Knowledge bases per practice** — Separate knowledge bases for tax, corporate, employment — each agent masters its practice area with citations. - **Access control** — Open to the public, or login-required for client-only agents — you choose per agent. - **HTTP tools** — Create enquiries in your practice-management system the moment intake completes. - **MCP server** — Expose firm tools over MCP so your team's Claude or Cursor can use them inside their daily workflow. - **Monetization** — Charge for premium agents with checkout, seats and payouts handled by the platform. ### FAQ **Can we keep an agent private to clients only?** Yes — per-agent access modes include login-required, and internal agents can stay inside your workspace entirely. **How do we stop the agent giving actual legal or tax advice?** You define its boundaries in plain language: answer process and general questions from approved content, and route anything that constitutes advice to your team. **Can we bill clients for access?** Yes — publish an agent with a monthly price and the platform handles subscriptions, seats and payouts. ## Travel agencies & tours https://supercognit.com/industries/travel-agencies Travel is sold across time zones: your customers plan at midnight and message from abroad. A SuperCognit agent trained on your itineraries, destination guides and policies answers instantly in 12 languages, qualifies the trip enquiry, and hands your consultants a lead with dates, party size and budget already captured. ### Use cases - **Itinerary & destination Q&A** — What's included, visa requirements, best season, difficulty level — answered from your own trip pages with citations, at any hour. - **Trip enquiry qualification** — Dates, travelers, budget and interests collected conversationally, then delivered to your consultants as a structured enquiry. - **Pre-departure support** — Packing lists, meeting points, transfer times and travel-document reminders — the questions that flood your inbox the week before departure. - **In-trip assistance** — Travelers message from the road on WhatsApp; the same agent answers with the itinerary knowledge and escalates real problems to your team. ### How SuperCognit maps onto Travel agencies & tours - **Website import → knowledge** — Your trip pages, FAQs and destination guides become a cited knowledge base in minutes, re-synced on a schedule as itineraries and prices change. - **WhatsApp & Telegram channels** — Travelers live on messaging apps — abroad especially. The same agent answers on both, in the traveler's language, with every conversation logged. - **HTTP tools → your booking system** — Check availability or create bookings and enquiries in your reservation system directly from the conversation. - **Widget + custom domain** — One script tag on your site, or a dedicated branded assistant at trips.youragency.com with SSL handled. - **Inbox & CRM** — Every enquiry lands with contact details and trip context, so consultants pick up warm, not cold. ### FAQ **Can the agent handle enquiries in different languages and time zones?** Yes — it answers in 12 languages automatically, around the clock. That's precisely where a human consultant team can't scale. **Can it check live availability?** Yes. Connect your booking or reservation API with the HTTP tool builder and the agent can check availability and create enquiries mid-conversation. **How does it stay accurate when itineraries change?** Imported pages re-sync on a schedule, so updated dates, prices and inclusions flow into the knowledge base without manual retraining. ## Restaurants & food service https://supercognit.com/industries/restaurants Every unanswered call at service time is a lost table. A SuperCognit agent trained on your menus, hours and policies answers the questions your staff can't get to — allergens, group bookings, parking, tasting menus — instantly, in the guest's language, on your website and on WhatsApp. ### Use cases - **Menu & allergen Q&A** — Ingredients, allergens, vegan options, wine pairings — answered accurately from your own menus, with citations. - **Reservation capture** — Date, time, party size and occasion collected in conversation, then pushed to your reservation system or booked via your booking link. - **Events & group enquiries** — Private-dining and celebration requests qualified with headcount, budget and date before your events manager gets involved. - **Practical questions, deflected** — Hours, parking, dress code, kids menu, terrace availability — the phone calls that interrupt service, answered without a human. ### How SuperCognit maps onto Restaurants & food service - **Website & menu import** — Menus, FAQs and event pages become the knowledge base; scheduled re-syncs keep seasonal menus current. - **Widget on your site** — One script tag adds the assistant to your site, styled to your brand, preloaded and instant. - **WhatsApp channel** — Guests already message restaurants on WhatsApp. The same agent answers there — menus, directions, bookings — and logs every conversation. - **HTTP tools → reservations** — Connect your reservation platform's API so the agent can create bookings or enquiries directly. - **Multilingual by default** — Tourist traffic asks in their own language; the agent answers in 11 of them, automatically. ### FAQ **Can the agent take reservations?** It can capture the full booking request in conversation and either push it into your reservation system via the HTTP tool builder or hand the guest your booking link at the right moment. **How do we keep allergen answers safe?** The agent answers from your own uploaded menus with citations, and you set its boundaries in plain language — including deferring uncertain allergen questions to staff. **Is this worth it for a single restaurant?** Setup takes an afternoon: import your site, review the draft, publish the widget. If it saves a handful of missed bookings a week, it has paid for itself. ## Fitness & wellness https://supercognit.com/industries/fitness-wellness Fitness and wellness run on momentum — the person asking about memberships at 11pm is ready to commit now, not tomorrow. A SuperCognit agent trained on your schedules, pricing and programs answers instantly, books the trial, and keeps members engaged between visits. ### Use cases - **Membership & pricing Q&A** — Plans, class packs, freezes, cancellation policy — answered clearly and consistently from your own materials. - **Trial & tour capture** — The agent qualifies goals and schedule, then books the trial class or tour — striking while motivation is hot. - **Schedule & availability** — Class times, instructors, spa treatment menus and prep instructions — the daily question load, deflected. - **Paid coaching companion** — Coaches publish an agent trained on their method — programs, form cues, nutrition guidelines — as a subscription product clients pay for monthly. ### How SuperCognit maps onto Fitness & wellness - **Knowledge from your programs** — Upload class descriptions, treatment menus, pricing and method guides; the agent answers with citations to your own content. - **Monetization built in** — Sell access to a premium coaching agent with checkout, seats and payouts handled by the platform. - **WhatsApp & Telegram** — Members message where they already are; the same agent answers on both channels. - **HTTP tools → booking** — Connect your class-booking or scheduling system so the agent books trials and sessions in conversation. - **Memory** — Returning clients are remembered, so the companion picks up their goals and history instead of starting over. ### FAQ **Can I charge clients for access to my coaching agent?** Yes — monetization is native. Set a monthly price and SuperCognit handles checkout, seats and payouts; keep it free for members if you prefer. **Will it give safe fitness advice?** It answers from your approved programs and guidelines, and you define its boundaries in plain language — including deferring medical questions to professionals. **Can it book classes and trials?** Yes, via the HTTP tool builder connected to your booking system, or by handing clients your booking link with details pre-qualified. ## SaaS & software https://supercognit.com/industries/saas SaaS buyers self-serve: they read your docs, test the product, and only talk to you when something blocks them. A SuperCognit agent trained on your documentation, changelog and pricing answers pre-sales and support questions instantly with citations — and because every workspace ships an MCP server, your product's knowledge and tools plug straight into the AI assistants your users already work in. ### Use cases - **Docs-grounded support** — Setup, integrations, limits, troubleshooting — answered from your documentation with citations, deflecting tickets before they're filed. - **Pre-sales qualification** — Feature, plan and compliance questions answered instantly; serious evaluators identified and routed to sales with context. - **Onboarding companion** — New users ask 'how do I…' in their own words and get the exact steps and doc links, cutting time-to-value. - **MCP distribution** — Expose your product's knowledge and API as an MCP server, so users can work with your product from Claude, Cursor and any MCP client. ### How SuperCognit maps onto SaaS & software - **Docs & changelog import** — Import your documentation site with scheduled re-syncs — the agent answers from the current version, not last quarter's. - **MCP server per workspace** — Your knowledge and HTTP tools become an MCP server your customers connect to from their own AI tools — a distribution channel, not just support. - **HTTP tools → your API** — Let the agent check account status, usage or plan limits through your own API, safely scoped. - **Widget + custom domain** — In-app help via the widget, plus a public assistant at ask.yourproduct.com. - **Cost caps** — Per-seat and global caps keep a launch-day traffic spike from becoming a bill spike. ### FAQ **How is this different from a docs-search widget?** The agent doesn't just find pages — it answers with citations, executes multi-step reasoning across docs, and can call your API to act. Plus the same knowledge ships as an MCP server for Claude and Cursor users. **Does it stay current with our releases?** Yes — imported docs re-sync on a schedule, so new features and changelog entries flow into answers automatically. **Can we keep it for logged-in customers only?** Yes — per-agent access modes include login-required, and you can run separate public (pre-sales) and private (customer) agents from one workspace. ## Sales teams https://supercognit.com/industries/sales-teams Most sales tooling watches reps after the fact. A SuperCognit agent works the earlier, cheaper moments: it plays the prospect so reps can practice objections against your actual playbook, it qualifies inbound conversations on your website and WhatsApp before a rep spends a minute, and it answers the everyday questions — pricing rules, discount limits, competitor talk tracks — from content you approved. Leads land in HubSpot or Pipedrive as contacts and deals with the full conversation attached. ### Use cases - **Objection roleplay** — An agent prompted as your buyer persona pushes back with the objections your team actually hears — reps practice the response out loud, on demand, without burning a manager's afternoon. - **Inbound qualification** — Website and channel conversations are qualified against your criteria; real prospects become CRM contacts and deals with the transcript attached, tyre-kickers get helpful answers and no follow-up task. - **Rep knowledge assistant** — "What's the EDU discount cap?" "How do we answer the security question?" — answered from approved playbooks and battle cards, consistently, instead of from whoever is online. - **Faster onboarding** — New reps ask the agent what they'd otherwise ask their manager, and drill the pitch against it until it sounds right — weeks of ramp compressed into days. ### How SuperCognit maps onto Sales teams - **Knowledge from your playbook** — Upload playbooks, battle cards and pricing docs, or import your wiki — answers cite the source, so enablement content finally gets used. - **HubSpot & Pipedrive built in** — Qualified leads are written as contacts and deals in the pipeline and stage you choose, matched on email, conversation attached as a note. - **Guardrails + Inbox** — You define what the agent may promise; every conversation is reviewable in the Inbox, so coaching happens on real transcripts. - **Login-required internal agents** — Roleplay and knowledge agents can be gated to signed-in team members, while the qualification agent stays public. - **Every channel reps live in** — Website widget for inbound, WhatsApp and Telegram for the rest — one agent, one CRM, one transcript trail. ### FAQ **Does this replace our sales trainer or enablement lead?** No — it multiplies them. Trainers set the playbook and review transcripts; the agent gives reps unlimited practice reps between sessions, which is exactly the part human coaching never scales to. **How does the CRM hand-off work?** Connect HubSpot or Pipedrive with OAuth, pick the pipeline, stage and owner, and map your fields. The agent matches contacts on email (no duplicates), opens the deal, and attaches the full conversation as a note. **Can the roleplay agent stay internal?** Yes — set it to login-required so only your team can use it, and keep a separate public agent for inbound qualification. Both live in the same workspace and share knowledge you choose. # Changelog https://supercognit.com/changelog Public release notes, newest first. ## One platform MCP connection per workspace 2026-09-29 · improved Each workspace now has its own platform MCP URL, so you can keep several workspaces connected in Claude, Cursor or any MCP client at the same time instead of disconnecting and reconnecting. - The URL is mcp.supercognit.com/platform//mcp — find yours under Integrations → Platform MCP, with ready-made Claude Code and mcp.json snippets named after the workspace. - A key from another workspace is refused on that URL, so a connection can never operate the wrong workspace by mistake. - The shared /platform/mcp keeps working for a single connection. ## Invite links, resend and a clearer Team tab 2026-09-28 · improved Every workspace invitation is now a link you can share on WhatsApp or Slack, and the Team tab shows when each invite was sent, lets you resend it, and keeps revoked invites with their date. - Copy link on every pending invite: whoever opens it joins the workspace with the role you chose, signed in or after creating an account. - Sent and resent dates on each invite, and a Resend button (at most once a minute). - Revoked invites stay listed with the date they were withdrawn, with an Invite again button. - The invitation email now opens the invitation directly instead of the login page. - Also in the API and MCP: /v1/invites (list, create, resend, revoke) and the list_invites, invite_member, resend_invite and revoke_invite platform tools. ## Saved replies, team presence and automatic assignment 2026-09-25 · new The last pieces for running live support from SuperCognit: canned answers on a slash, an Available / Away switch, new hand-overs assigned to whoever is online, and an Inbox that updates itself. - Saved replies under Settings → Saved replies. Type / in the reply box to pick one; {{name}} becomes the visitor's first name and {{agent}} yours. - An Available / Away switch in the Inbox, and a dot next to every teammate who is online. - New hand-overs go to the least busy teammate who is online, and only that person is alerted. Turn it off per agent to let the team pick. - The Inbox list refreshes itself, so new and waiting conversations appear without a reload. - A visitor on the web who asks for a person is asked for an email. If they have left by the time you reply, the answer is also emailed to them with a link back to the chat. - With human support on, the widget header shows whether your team is online or the AI is answering on its own. - Also in the API and MCP: /v1/saved-replies, presence on /v1/members, and saved-reply platform tools. ## Know when a visitor is waiting 2026-09-25 · new When a conversation needs a person, the right people hear about it: by email, with a badge on the Inbox and, if they want, a desktop notification. - An email when a visitor asks for a person, when a conversation is assigned to you, and when the visitor writes again while waiting — at most one every 10 minutes per conversation. - Owners, admins and support agents hear about every hand-over by default; people who build agents only about their own. Everyone sets their own under Inbox → Alerts. - A badge on the Inbox counts the conversations waiting for a person and yours where the visitor wrote last. - Optional desktop notifications, turned on per browser. - Four new webhooks for your own tools: conversation.handoff_requested, conversation.assigned, conversation.team_replied and conversation.resolved. ## Hand the conversation to a person 2026-09-25 · new Turn on human support for an agent and it hands the conversation to your team when a visitor asks for a person or when it cannot help. While someone from the team is on it, the agent stays quiet. - A new Support tab in the agent editor: turn it on, choose when the agent hands over, and set your business hours. - A "Talk to a person" button on the chat page and in the widget — always, after the agent's first answer, or never. - The agent hands over by itself when the visitor asks for a person or when it is stuck, and tells the visitor the team is on its way. - While a person holds the conversation the agent does not answer; messages wait for your team, on the web, WhatsApp and Telegram alike. - Outside business hours you choose: the agent keeps helping, or the conversation waits in the queue for the team. - If nobody from the team replies within 10 minutes the agent comes back and carries on, so no one is left waiting. Set anything from 2 to 120 minutes. - Configurable from your own code as humanSupport on /v1/products and the update_agent platform tool. ## Reply to visitors from the inbox 2026-09-25 · new Your team can now answer a conversation from the Inbox. The reply reaches the visitor wherever they are talking — the chat page, the widget, WhatsApp or Telegram — signed with the team member's first name. - A reply box under every conversation, with emoji, Enter to send and a note of which channel the answer goes out on. - Visitors see the team member's first name and photo next to the message, marked as someone from your team — never their email address. - An open chat page or widget shows the reply within seconds, without a reload. - On WhatsApp, Meta only allows free-form replies within 24 hours of the customer's last message; after that the reply box tells you instead of failing silently. - Whoever replies first becomes the conversation's owner if nobody had it yet. - Also in the API and MCP: POST /v1/sessions/{id}/reply and the reply_to_conversation platform tool. ## Leads without the interrogation 2026-09-24 · new Agents now offer to take a visitor's details at the right moment — after answering, once someone asks about prices, availability or a quote — and collect them through a short form with a consent checkbox instead of spelling out an email in the conversation. - A per-agent policy: when the visitor shows interest (the default for new agents), early, form first, or off. - A contact card on the chat page and in the widget: name, email and the questions you configure, with a consent checkbox. On WhatsApp and Telegram the agent asks in the conversation instead. - New agents get a lead tool automatically; leads land under Leads and in your CRM if one is connected, with the page they came from. - You get an email for every new lead, and visitors who left their details are no longer limited to three conversations. - Sites that already know the visitor can call SupercognitWidget.identify({ email, name }) — signed on your server to mark it verified — and the agent never asks again. - Available from your own code too: leadCapture on the platform MCP tools and on POST/PATCH /v1/products. ## WhatsApp without paying for every reply 2026-09-24 · new From 1 October Meta bills every reply an agent sends on WhatsApp once a number has used its free monthly allowance. A new cost control answers up to a budget you set, then moves the person to your web chat with the conversation intact. - Off by default. Turn it on per agent under Channels → WhatsApp cost control; if you are happy to pay Meta, nothing changes. - Two budgets: answer for a number of hours from the person's first message, and a monthly cap on delivered messages per number — 1,000 matches Meta's free allowance. - Past the budget the agent sends a short reminder with a private link instead of an answer, a fixed number of times, then goes quiet. A week of silence starts the person fresh. - The link opens your web chat with the WhatsApp conversation already there, so either side can carry on. In the web chat it shows as a card. - The reminder comes in the agent's language, or write your own with a {{link}} placeholder. - Telegram is not affected. Also available as whatsappHandoff on the platform MCP tools and on /v1/products. ## Choose the countries your agent serves 2026-09-24 · new A country rule on the Guardrails tab blocks the countries you list, or serves only the ones you list, so an agent built for one market stops spending on visitors it will never serve. - Two modes: block the listed countries, or serve only the listed countries. Tor exit nodes can be listed too. - On the chat page a blocked visitor sees a short notice in their language and nothing about the agent; the widget simply does not appear. - Applies to the chat page, the widget and the chat API. WhatsApp, Telegram, the public API and your own preview are never affected. - Visitors whose country cannot be determined are always served — this is a cost and noise filter for markets you do not serve, not a security control. - Manage it from your own code: get_guardrails and set_guardrails on the platform MCP tools, and GET/PATCH /v1/products/{id}/guardrails. ## An inbox your team can work 2026-09-24 · new Conversations now carry a status, an assignee, a priority, tags and internal notes, and the Inbox has views for what needs a person — the groundwork for your team taking over from the agent. - Statuses: open, needs a person, with a person, resolved. A visitor writing into a resolved conversation reopens it. - Inbox views — All, Mine, Unassigned, Needs a person, Resolved — with counts, a tag filter, and status, assignee and priority on every row. - Take a conversation, hand it to a colleague, set its priority, tag it and leave internal notes the visitor never sees; every change is in the conversation's history. - Tags are managed under Settings → Tags, with usage counts. - A new agent role for support staff: inbox only, no access to agent building, billing or settings. Seats count against your plan. - In the API and MCP too: /v1/sessions filters and workflow, notes, conversation tags and members, plus nine new platform MCP tools. - Coming next: replying from the inbox, and the agent staying quiet while a person handles the conversation. ## Analytics: who visits, who talks, who converts 2026-09-24 · new A new Analytics section in the dashboard shows how people find and use your agents — visits, conversations, messages and leads over time, the busiest hours, and a map of where they come from. Pro plans get a live map of who is on the site right now. - Overview, Conversations, Visitors and Leads pages with a 24h / 7d / 30d / 90d range, an agent filter and your own timezone. - A world map of visits, conversations or leads by country and city, from coarse network location — never a precise position, never an IP. - Busiest hours as a weekday × hour grid, so you know when to staff and when to let the agent handle it. - Pro and Enterprise: the live map of visitors online and chatting now, mean response times, the lead funnel through to CRM delivery, devices, browsers, languages, previous-period comparison and CSV export. - Tools and MCP servers too: calls per tool with outcomes, durations and cost share, and how each MCP server you publish is used by external clients. - Everything is also in the API (/v1/analytics/…) and in the platform MCP tools, so your own dashboards and assistants can ask the same questions. ## Options as cards, not paragraphs 2026-09-15 · new When the agent has a handful of things to show — rooms, products, plans, appointments — it now lays them out as cards with an image, a title and a button, and the visitor's tap flows straight back into the conversation. - Two to ten options as a card strip or grid: image, title, subtitle, a price or badge, and a button that posts the choice back into the chat. - A "More details" sheet on every card, with a gallery, address, phone, email, hours and link, plus longer sections. - Works on the chat page and in the widget. On WhatsApp cards go out as an interactive carousel, on Telegram as photo cards with buttons — a tap becomes the same choice. - Card images only ever come from data a tool returned in that turn, so the agent can't make one up. ## The agent plays your videos 2026-09-15 · new Add YouTube or Vimeo videos to an agent's Media tab and it plays the right one inline when a visitor asks for a tour, an intro or a demo — in our own player, not the native embed. - Up to 25 videos per agent, added from the new Media tab in the editor. - Plays inline on the chat page and in the widget: a click-to-play poster, our controls and your agent's accent colour. - On WhatsApp and Telegram the agent sends the video's title and link instead. - Manage media from your own code too, through the platform MCP tools and /v1/agents/{id}/media. ## Website import works behind bot challenges 2026-09-15 · improved Some sites greet crawlers with a challenge page instead of content. Import now reads those through our own Chromium, so building an agent from a protected site no longer fails at the first step. - Sites behind bot challenges are fetched with a real browser and imported like any other. - Nothing to configure — the importer picks the route on its own. - Same agent, same knowledge base, same cited answers as a plain import. ## Better answers from big knowledge bases, fresher on a schedule 2026-09-14 · improved Retrieval now adapts to the size of your knowledge, keyword search matches exact names again, and scheduled re-syncs only redo the pages that actually changed. - Retrieval sizes itself to the corpus instead of always pulling a fixed five passages, so large knowledge bases stop dropping the right page. - The keyword arm of hybrid search was repaired: exact product names, SKUs and codes match again. - Scheduled refresh skips pages whose content hasn't changed, so a re-sync is cheaper and faster. - Refresh cadence is set per source — daily on Starter, every 6 hours on Pro, hourly on Enterprise. ## Plans now include usage credits 2026-09-11 · improved Every paid plan comes with a monthly allowance of usage credits, so the subscription covers real conversations from day one instead of billing every reply separately. - Starter and Pro each include a monthly allowance of usage credits (amounts on the pricing page); the 7-day trial carries a quarter of it. - Plan credits reset at renewal; credits you buy as a top-up never expire. - Switching between Starter and Pro changes your existing subscription in place, prorated, and keeps a running trial. ## A portal for partners 2026-08-28 · new Partners now have one place for everything client-facing: the workspaces you manage, your firm's details and brand, the sales deck and your certifications. - Clients, Details, Brand, Presentations and Certifications — the whole partner relationship in one portal. - Put your logo and colours on the sales deck and export it to PDF to send ahead of a meeting. - Work inside every client workspace from the dashboard login you already have — no separate accounts. ## Turn conversations into leads — with or without a CRM 2026-08-27 · new Your agent can ask for the details that matter and file the answer where you already work: straight into Pipedrive or HubSpot, or into SuperCognit itself if you don't run a CRM. - Connect Pipedrive or HubSpot and a captured lead lands as a real deal, in the right pipeline and stage. - No CRM? We keep the leads for you — visible in the inbox, listed on their own page, and pushed onward by webhook or API. - Change what a capture asks for whenever you like, without rebuilding the tool or detaching it from your agents. - If your CRM is unreachable, leads wait in line and go through the moment it answers again. ## The API reference is public 2026-08-24 · new Everything the dashboard does, your code can do too — and the full reference now lives online, interactive and open to everyone, next to a new page explaining what the API covers. - Browse every endpoint at api.supercognit.com/docs and try calls straight from the page. - Create, publish and update agents and MCP servers, tools and knowledge from your own code. - Receive conversations as they happen through signed webhooks, or read the transcripts later. ## Connect SuperCognit to Claude, Cursor and other AI clients 2026-08-16 · new Your workspace now speaks MCP in both directions: as well as publishing your own MCP servers, you can plug SuperCognit itself into the AI client you already work in. - Create and edit agents, knowledge bases and tools straight from your assistant. - Read your inbox and your contacts without leaving the client you work in. - Works with any MCP-capable client, Claude and Cursor included. ## Agents that use a real browser 2026-08-15 · new The new browser tool lets an agent open a site, fill in a form and bring the answer back into the conversation — for the jobs that live behind a portal instead of an API. - Check availability, submit forms and collect documents from sites with no API. - Longer jobs run in the background, and the agent tells the customer the moment they finish. - Results come back as text or as a PDF, ready to use in the chat. ## WhatsApp and Telegram channels 2026-08-14 · new Your agent now answers where your customers already are. Connect a WhatsApp Business number or a Telegram bot and the same agent picks up the conversation. - One agent, every channel — same prompt, same knowledge, same tools. - Conversations from every channel land in the shared inbox. - Customers get an answer in their own language, whichever channel they wrote from. ## withnodes is now SuperCognit 2026-08-13 · improved New name, new site, same platform. The rebrand landed across the product, together with a redesigned homepage and a site that finally speaks more than one language. - Everything you built keeps working — agents, domains and integrations are untouched. - The site now ships in English, Portuguese and Spanish. ## From your website to a working agent in two minutes 2026-07-13 · new Paste your address and we read your public pages, draft the prompt and build the first knowledge base for you. You review it, adjust the tone and publish. - The import wizard takes you from a URL to a live draft agent. - Your pages become knowledge sources you can edit, extend or remove. - Also available through the API, for agencies onboarding clients in bulk. ## Public or sign-in-only agents 2026-07-13 · improved Every agent now has an access mode, so a pre-sales assistant and a customers-only one can live side by side in the same workspace. - Open: anyone with the link can start chatting. - Sign-in required: visitors log in before the first message. - Widgets embedded on your own site keep working either way. ## Inbox and contacts 2026-07-12 · new Every conversation your agents have, in one place — with a contact record behind each one, so you can see what a customer asked before they wrote again. - Read full transcripts from every channel in a single inbox. - Contacts gather the conversations of each person who wrote to you. - Knowledge and tools moved into a cleaner home in the dashboard at the same time. ## Simpler plans, priced in euros 2026-07-10 · improved Two plans — Starter and Pro — each with a seven-day free trial, and a much clearer picture of what a published agent costs to run. - Choose a plan and start the trial straight from the pricing page. - Publishing an agent now tells you upfront what running it will cost. ## Embeddable widget, webhooks and a bigger API 2026-07-10 · new Three ways to take your agent off our pages and put it inside your own product. - Drop one script tag on your site for a chat bubble that opens instantly. - Webhooks push conversation events into your own systems. - The public API covers agents, knowledge and conversations. ## Agents that answer in your customer's language 2026-07-10 · new Publish once and your agent replies in twelve languages out of the box, following whoever wrote in. - Names, descriptions and greetings can be translated language by language. - The customer never has to pick a language — the agent simply follows theirs. ## Faster answers, web search and deeper thinking 2026-07-08 · improved A round of model-side work made agents quicker on repeat questions and noticeably better on hard ones. - Repeated context is reused between messages, so replies start sooner and cost less to run. - Agents can search the web when your own knowledge doesn't cover the question. - Harder questions get more thinking time, and long conversations keep their thread. ## Launch on your own domain 2026-07-07 · new Your agent can live at your address instead of ours. Connect a domain and the certificate and routing are handled for you. - Point a subdomain such as chat.yourbrand.com at your agent. - Set it up in the dashboard; it goes live as soon as the DNS records match. ## Bring your own knowledge 2026-07-04 · new Agents can be grounded in your documents, not just in your prompt — and they show where an answer came from. - Upload PDFs, Word documents, text and markdown files. - Add website pages and Google Docs links that re-sync on a schedule. - Answers cite the source they were built from. ## Charge for your agents 2026-07-04 · new Turn a published agent into a product: set a price, and subscriptions, invoices and payouts are handled for you. - Per-seat subscriptions for the teams that buy your agent. - Checkout, invoicing and refunds run on the platform. - A clear breakdown of what you earn on every subscription. ## Partner program 2026-07-02 · new Agencies and consultancies can build on SuperCognit for their own clients, at wholesale pricing. - A partner portal for the client workspaces you run. - Wholesale rates on agent seats and MCP servers. ## Agents that do things, not just answer 2026-07-01 · new Agents run a multi-step tool loop: they call your APIs, search your knowledge and use your MCP servers to finish a job inside a single conversation. - Connect any HTTP API as a tool, no code required. - Publish your workspace's tools as an MCP server other clients can connect to. - Shared tool and knowledge libraries, so several agents reuse the same setup. # Blog ## Your team can now take over from the AI agent https://supercognit.com/blog/ai-agent-human-handover · 2026-09-25 SuperCognit agents can now hand a conversation to a person. Your team replies from the Inbox on the web, WhatsApp and Telegram, the agent stays quiet while they do, and it comes back if nobody answers. Until this week, an agent that could not help had one move left: point the visitor somewhere else. The conversation landed in the Inbox and someone could follow up later by email. That works for a lead. It does not work for a customer who is on your site right now with a question the agent cannot answer. Now a person from your team can step into the conversation, answer in the same chat and hand it back. Here is what that looks like and how to turn it on. ### How a hand-over works 1. **The visitor asks, or the agent gets stuck** — Visitors can press "Talk to a person" in the chat or the widget, or simply ask. The agent also hands over by itself when it cannot help. 2. **The right person hears about it** — The conversation goes to the least busy teammate who is online. They get an email, a badge on the Inbox and, if they want, a desktop notification. 3. **Your team replies, the agent waits** — The reply reaches the visitor where they are talking, signed with the teammate's first name. While a person holds the conversation, the agent does not answer. 4. **Resolve it, or let the agent carry on** — Mark it resolved when it is done. If nobody replies within the window you set, the agent comes back and keeps helping, so no one is left waiting. ### Everything a support team needs, in the Inbox The Inbox already had statuses, assignment, priority, tags and internal notes. This release adds the parts that make it work for live support: - A reply box on every conversation, with emoji and saved replies: type / and a shortcut, and {{name}} fills in the visitor's first name. - An Available / Away switch, and a dot next to every teammate who is online. - A list that refreshes itself, so waiting conversations appear without a reload. - An Agent role for support staff: the Inbox only, with no access to agent building, billing or settings. - Webhooks for hand-over requests, assignments, team replies and resolutions, for teams that run on their own tools. ### On WhatsApp and Telegram too A reply from the Inbox is delivered in the same WhatsApp or Telegram chat the customer started, not in a new thread. One Meta rule applies: free-form WhatsApp replies are only allowed within 24 hours of the customer's last message, and the reply box tells you when that window has closed. ### When the visitor has left People do not wait on a web page. A visitor who asks for a person on the web is asked for an email address, and if they have closed the tab by the time your reply arrives, the reply is also emailed to them with a link back to the conversation. ### Outside business hours Set your hours per agent, in your time zone. Outside them you choose whether the agent keeps helping on its own or the conversation waits in the queue for the team. Either way the visitor is told what to expect. ### Turn it on 1. **Open the Support tab** — In the agent editor, turn on human support, choose when the agent hands over and set your hours. 2. **Invite your team** — Add teammates under Settings → Team. Give support staff the Agent role; the number of seats depends on your plan. 3. **Add a few saved replies** — Greetings, opening hours, the refund policy: the answers your team types every day, under Settings → Saved replies. Everything here is also in the API and the platform MCP tools: replies, hand-over settings, assignment, tags, notes and saved replies. The changelog has the details of each part. ## What's new: WhatsApp cost control, leads without the interrogation, country rules and a team inbox https://supercognit.com/blog/whats-new-late-september-2026 · 2026-09-24 Ten days after the last update: a WhatsApp cost control to have ready before Meta's 1 October pricing change, a contact card that captures leads without spelling out an email in the chat, a country rule for markets you do not serve, and the first step towards people taking over from the agent. Ten days after the last update, four things you will notice and one you will want to turn on before 1 October. Here is the short version; each item has its own changelog entry with the details. ### WhatsApp: a budget instead of a bill On 1 October 2026 Meta starts billing every message a business sends on WhatsApp — including the free-form replies inside the 24-hour window — once a number has used its 1,000 free messages for the month. For an agent that handles a few thousand conversations a month that is a new line on the bill, and it grows with success. Two things to know before then: numbers without a payment method on the WhatsApp Business Account stop delivering replies on that date, so add one by 30 September; and the platform now has a cost control for it. The control is off by default and lives on the agent's Channels tab. When it is on, the agent answers up to a budget you set — a number of hours from the person's first message, a monthly cap on delivered messages per number, or both — and past that it sends a short reminder with a private link instead of an answer, a fixed number of times, then goes quiet. The link opens your web chat with the WhatsApp conversation already there, so nothing is lost and either side can carry on. In the web chat the reminder shows as a card, not a bare URL. - Defaults that match Meta: 24 hours and 1,000 messages a month, five reminders. Change any of them, or turn a rule off. - The reminder comes in the agent's language; write your own with a {{link}} placeholder if you prefer. - Telegram is free and untouched. The setting is also on the platform MCP tools and /v1/products as whatsappHandoff. ### Leads without the interrogation Until now a lead existed only when you built a CRM tool and prompted the agent to ask for details. Agents now offer to take a visitor's details at the right moment — after answering, once someone asks about prices, availability or a quote — through a short contact card with a consent checkbox, on the chat page and in the widget. On WhatsApp and Telegram the agent asks in the conversation instead. - A per-agent policy: when the visitor shows interest (the default for new agents), early, form first, or off. Existing agents stay off until you choose. - Leads land under Leads and in your CRM if one is connected, with the page they came from, and you get an email for each one. - Sites that already know the visitor can call SupercognitWidget.identify({ email, name }) — signed on your server to mark it verified — and the agent never asks again. - Also on the platform MCP tools and POST/PATCH /v1/products as leadCapture. ### Country rules An agent built for one market used to answer anyone who found it. The Guardrails tab now has a country rule with two modes: block the countries you list, or serve only the ones you list. A blocked visitor sees a short notice in their language and nothing about the agent; the widget simply does not appear; the chat API refuses before any usage is spent. WhatsApp, Telegram, the public API and your own preview are never affected, and visitors whose country cannot be determined are always served — this is a cost and noise filter, not a security control. ### A team inbox, first step Conversations now carry a status — open, needs a person, with a person, resolved — plus an assignee, a priority, tags and internal notes the visitor never sees. The Inbox has views for All, Mine, Unassigned, Needs a person and Resolved, with counts and a tag filter, and every change is in the conversation's history. There is a new agent role for support staff that only reaches the inbox, and tags are managed under Settings. This is the groundwork: replying from the inbox and the agent staying quiet while a person handles a conversation are the next two releases. ### Also new - Analytics: an Overview, Conversations, Visitors and Leads section in the dashboard, with a world map of where people come from and the busiest hours. Pro plans get a live map of who is on the site right now. - Everything above is in the API and the platform MCP tools as well, so your own dashboards and assistants can use it. Everything here is live today. The changelog at supercognit.com/changelog has one entry per item, and that is where we keep writing these up. ## An MCP server behaved for three calls, then went looking for your SSH keys https://supercognit.com/blog/mcp-tool-poisoning-deadbugz · 2026-09-21 Security researchers disclosed an active campaign — tracked as Deadbugz — distributing a malicious MCP server through GitHub pull requests. It answers normally for two tool calls, then rewrites its own tool metadata on the third to hunt for credentials. Nothing was compromised, but the technique is the story. We have written before about MCP servers nobody vetted and about the majority of public servers that still skip OAuth. Both are about servers you should not have trusted in the first place. Deadbugz, an active campaign that researchers at Pillar Security disclosed and that the community MCP incident tracker logged as INC-136 on 18 September 2026, is a different problem: a server that looked fine when you approved it, and changed what it does afterward. ### What was found A single GitHub account filed 23 pull requests across unrelated AI and developer-tooling repositories, each one adding a small MCP server pitched as a productivity utility — text formatting and summarization, nothing that would raise an eyebrow in a code review. None of the pull requests were merged before the campaign was caught, and no compromise has been confirmed as of disclosure; researchers are treating this as a demonstrated technique, not a breach with victims. ### The mechanism: call-count gating The server's behaviour depends on how many times a client has called it, not on what it is asked to do — which is what let it pass a quick look and a first session unnoticed. 1. **Calls one and two** — The server returns exactly what its description promises: formatted or summarized text. A reviewer testing it sees a small, boring, working tool. 2. **Call three** — The server rewrites the metadata it returns — the tool descriptions the model reads as instructions — into directions to search for SSH keys, AWS credentials, shell history and Kubernetes config, and to do it quietly. 3. **Why session-based review misses it** — Most manual testing and most one-off approval flows stop after a call or two. The payload is timed to appear after the point where a human has usually stopped watching. ### Why this is a distinct threat from a server that was malicious on day one Tool descriptions in MCP are not documentation for a human — they are text the connected model reads and can act on, which is what makes prompt injection via tool metadata possible at all. A one-time approval assumes the thing you approved stays the thing that is running. Deadbugz demonstrates that assumption is not currently enforced anywhere in the ecosystem: nothing in the specification requires a signed tool manifest, and nothing in most clients re-prompts for consent when a previously approved server's tool definitions change. ### What to actually do about it - Fingerprint tool definitions at approval time, and compare the fingerprint on every reconnect — a change is a security event, not a routine update. - Treat any definition change on an already-approved server as requiring renewed, explicit consent, not silent auto-acceptance. - Do not rely on a first session to vet a third-party server. If your process is 'try it once, then trust it,' this is exactly the gap that exploits. - Prefer servers you or your platform control the definition of over ones pulled from an open registry or a pull request you did not write. On SuperCognit, an agent's tools are HTTP endpoints the workspace owner defines and can see in full, not third-party servers pulled in from a registry — there is no pull request from a stranger in that path. That sidesteps this specific campaign, but it does not make the underlying problem go away: anyone connecting Claude, Claude Code or any other MCP client to servers from the open ecosystem should treat the source of a server, and any change in what it advertises, as things worth actually checking. This account is based on Pillar Security's public disclosure and the community MCP incident database entry (INC-136); some detail is drawn from secondary reporting rather than the original write-up, and neither source has published a confirmed victim as of this writing. ## What's new: cards, video and included usage https://supercognit.com/blog/whats-new-september-2026 · 2026-09-15 September's release: agents show options as cards and play your videos inline, paid plans now include monthly usage credits, and a batch of smaller things — protected-site import, better retrieval on large knowledge bases, a partner portal. Since the last update in late August we've shipped three things you'll notice in the first conversation and a handful you'll notice over the month. Here is the short version; each item has its own changelog entry with the details. ### Cards: options you can tap Ask an agent about rooms, plans or products and until now you got a paragraph. It now answers with cards — two to ten options in a strip or a grid, each with an image, a title, a subtitle, a price or badge, and a button. Tapping the button posts the choice back into the conversation, so "the second one" stops being a guess. A "More details" sheet on each card carries a gallery, facts such as address, phone, hours and link, and longer sections. - Works on the chat page and in the widget. - On WhatsApp cards go out as an interactive carousel, on Telegram as photo cards with inline buttons — a tap becomes the same choice. - Card images only ever come from data a tool returned in that turn; the agent cannot invent one. ### Video: the agent shows, not just tells Add YouTube or Vimeo videos to an agent's Media tab — up to 25 — and when a visitor asks for a tour, an intro or a demo the agent plays the matching one inline. It runs in our own player, with a click-to-play poster, our controls and the agent's accent colour, rather than the platform's native embed. On WhatsApp and Telegram the agent sends the title and the link instead. Media can also be managed through the platform MCP tools and /v1/agents/{id}/media. ### Plans now include usage credits Every paid plan now carries a monthly allowance of usage credits — the amounts are on the pricing page — and the 7-day trial comes with a quarter of it. Plan credits reset at renewal; top-ups you buy never expire. Switching between Starter and Pro changes the existing subscription in place, prorated, and keeps a running trial. ### Smaller things - Website import now works on sites behind bot challenges — they are read through our own Chromium instead of failing at the first step. - Retrieval on large knowledge bases sizes itself to the corpus instead of a fixed five passages, and the keyword arm of hybrid search was repaired, so exact names, SKUs and codes match again. - Scheduled knowledge refresh skips pages whose content hasn't changed. Cadence is per source: daily on Starter, every 6 hours on Pro, hourly on Enterprise. - Partners have a portal — Clients, Details, Brand, Presentations and Certifications — and the sales deck exports to PDF in your brand. - Chat tabs show the agent's favicon, workspace pickers show avatars, and your own preview of an agent no longer counts against guest limits. Everything above is live today. The changelog at supercognit.com/changelog has one entry per item, and that is where we'll keep writing these up. ## The first MCP vulnerability just landed on CISA's exploited list https://supercognit.com/blog/mcp-cisa-kev-vulnerabilities · 2026-09-14 On September 2, CISA added a LiteLLM authentication bypass to its Known Exploited Vulnerabilities catalog — the first Model Context Protocol flaw on that list. A critical bug in AWS Labs' own Postgres MCP server followed a week later. On September 2, 2026, CISA added a Model Context Protocol vulnerability to its Known Exploited Vulnerabilities catalog for the first time. Five days later a second, more severe bug turned up in a different, widely-used MCP server. Neither is an exotic MCP-specific flaw — one is an authentication fallback that failed open, the other is a database privilege bypass — which is exactly the point: MCP is deployed widely enough now that ordinary bug classes in ordinary infrastructure are what takes it down. ### The first MCP entry on CISA's exploited list CVE-2026-59822 is an authentication bypass in LiteLLM, the open-source proxy a large share of self-hosted AI stacks use to route calls to models and, for many teams, to their MCP servers. CISA added it to the Known Exploited Vulnerabilities catalog on September 2, with a remediation deadline of September 16 for US federal systems — a fast clock that, under the agency's new Binding Operational Directive 26-04, is reserved for flaws that are confirmed exploited, reachable from the public internet and easy to automate. As far as we can tell, it is the first Model Context Protocol implementation to appear on that list. ### How the bypass worked LiteLLM's MCP Streamable HTTP endpoint checks the caller's key and has a fallback path for upstream OAuth2 passthrough. When key validation failed, the fallback did not deny the request — it substituted an empty UserAPIKeyAuth() object and let the call through anyway. A request carrying nothing more than a fabricated Authorization header could reach whatever MCP tools were configured, no valid key required. The fix, shipped in version 1.84.0, closes that fallback. LiteLLM published the advisory back in June; CISA's listing two months later is what confirmed the flaw was actually being used against real deployments, not just theoretically exploitable. ### A second, more severe bug five days later On September 9, AWS Labs disclosed CVE-2026-87911 in its own postgres-mcp-server: a critical, CVSS 9.6 flaw where the SQL validation behind the server's read-only mode could be defeated with a crafted COPY ... TO PROGRAM statement, letting an attacker run operating-system commands on the database host. It only bites when the MCP server connects with a role that holds superuser or pg_execute_server_program privileges — a shortcut plenty of quick setups take, because it is the path of least resistance until something reads the advisory for you. Fixed in 1.1.7. ### Why this is landing on MCP specifically Neither bug is a flaw in the MCP specification — an auth fallback that fails open and a privilege escalation via COPY TO PROGRAM are bug classes as old as the technologies underneath them. What changed is that fixing them on an accelerated clock is now worth CISA's and AWS's time, because MCP servers sit exactly where a database or API gateway does: reachable, holding credentials, and now common enough to be a target class instead of a curiosity. CISA's September 2 batch logged seven exploited flaws; three, including this one, were reported as targeting AI infrastructure specifically. ### What to actually do about it - Patch LiteLLM to 1.84.0+ and postgres-mcp-server to 1.1.7+ if you run either — both fixes are already out. - Whatever MCP server touches a database, connect it with a role scoped to the tables and operations it actually needs, never superuser. Read-only mode is a safety net, not a substitute for database-level permissions. - Test the failure path of your auth, not just the success path. A validator that fails open is worse than one that crashes, and from the outside they look identical until someone finds the gap. - If you can't patch immediately, put the MCP proxy behind a gateway that can gate or rate-limit the /mcp/ routes on its own. It also underlines the point our own OAuth checklist has been making since August: the specification requires OAuth 2.1 on every MCP server, and plenty of servers still cut that corner. These two CVEs are what cutting it costs in practice — not a hypothetical, a dated CISA catalog entry with a remediation clock attached. ## How to sell MCP: a field guide for agencies, consultants and internal champions https://supercognit.com/blog/how-to-sell-mcp · 2026-09-10 Selling MCP means selling correct, cited answers and permitted actions inside the AI tools people already use — not the protocol. Here is how to pitch the benefits, package it by industry and department, sell it inside a company, and run MCP servers for many customers from one place. "Selling MCP" is shorthand for selling a company's knowledge and actions, made available to AI assistants in a form they can actually use: an MCP server the buyer's team connects from Claude or any other MCP client, or that the buyer's own customers connect to. Nobody buys a protocol. They buy answers that are right and cited, actions that are permitted rather than improvised, and the end of copy-and-paste. On SuperCognit a server is built without code from a website, documents and tools, published private or public, sold per seat if you want, and run alongside every other server you operate for every other customer. This guide covers the whole loop: what to sell, to whom, in which words, how to package it by industry and by department, how to sell it inside a company, and how to manage a portfolio of them. ### Why "MCP" is suddenly a search term The Model Context Protocol went from one vendor's specification to the shared plug standard for AI tools in under two years. Claude, coding assistants, desktop apps and a growing set of enterprise software all speak it; the spec is now stewarded by a foundation rather than a single company, and its auth model has already been through a major revision. Commercially that means one thing: the buyer's staff already have MCP clients on their machines. The question in the market has moved from "what is this" to "who will build ours" — which is why the phrase is being typed into search bars by agency owners, consultants and IT leads at the same time. The window matters. Every business that gets an MCP server this year gets it from somebody. Being that somebody is the opportunity, and the skill it takes is more commercial than technical: knowing what to promise, what to show, and what not to say. ### What you are actually selling Strip the acronym away and the offer has four parts. Sell these, in this order, and the protocol becomes a footnote. - Accuracy. Assistants improvise about companies they have never seen. An MCP server answers from published, approved material, with citations pointing at the source. The pitch is not "AI" — it is "the AI stops guessing about you". - Reach. The answers show up where the buyer's people already work: inside Claude, inside the assistants their customers use, and — from the same build — in a chat widget on the website, on WhatsApp and Telegram, and over an API. One build, many doors. - Permission. Tools turn a server from a reference into a colleague: check availability, look up an order, create a ticket, capture a lead. Each action is one the buyer chose to expose, and every user is authenticated — OAuth 2.1, per person, per server. - Maintenance you don't do. Knowledge re-syncs on a schedule, as often as hourly, so the server tracks the website and the documents instead of drifting from them. This is the line that turns a project into a subscription. ### Translating benefits into the buyer's words Buyers do not share your vocabulary, and the fastest way to lose a sale is to teach it to them. Say the outcome; let them ask about the mechanism. | What you might say | What the buyer hears | Say this instead | | --- | --- | --- | | We'll deploy an MCP server | A project with a server in it | Your team will get correct, cited answers about your business inside Claude | | It exposes tools over the protocol | Something my developers will have to maintain | The assistant can check availability and create a ticket — and only those two things | | It's grounded on your knowledge base | Another place to upload documents | It reads your website and the documents you already have, and follows them when they change | | It supports OAuth 2.1 and workspace isolation | Security words | Every person signs in. Nobody sees anything you didn't publish. Nothing is shared with anyone else's server. | | Per-seat SaaS pricing with a merchant of record | Complicated | Ten people costs this much a month. One invoice. Cancel any time. | ### The objections, and the honest answers Five objections come up in almost every conversation. Have the answers ready, and keep them short. 1. **"We already have a chatbot"** — The chatbot is one door; the server is the room behind every door. The same knowledge that powers the widget also answers inside Claude, on WhatsApp, and to any AI tool a customer points at your company. And it cites its sources, which most chatbots do not. 2. **"We already have an API"** — An API is for developers who write code against it. An MCP server is for assistants that read it. Nobody integrates; they connect. If the buyer has an API, that is good news: its endpoints become tools. 3. **"AI makes things up"** — On its own, yes. This server answers only from what you published, shows where each answer came from, and says so when it does not know. The three-questions test below is how you prove it in the room. 4. **"Our data changes all the time"** — That is the argument for the server, not against it. Sync runs on a schedule; the rate sheet updated on Tuesday is what the assistant answers with on Tuesday. 5. **"Is it secure?"** — Every server implements the current MCP auth spec — OAuth 2.1 with issuer validation and CIMD client registration — and lives in its own isolated workspace. A private server is behind login; a public one is a deliberate choice, made per server. Custom domains keep the address theirs. ### The demo that closes: build it before the meeting The single most effective sales move in this market is to arrive with the buyer's server already built. Import their website — it is crawled into a cited knowledge base in about two minutes — and you walk in with a chat agent that answers questions about their business better than their own homepage does. The first fifteen minutes of the meeting are the buyer asking it things. Then run the three-questions test: ask the three questions their team answers most often. If the answers come back right and cited, the sale is mostly done. If one comes back wrong, you have found the document they need to send you — which is the kickoff email written for you. Either outcome moves the deal. The chat agent is the demo; the MCP server is the deliverable. They are the same build, so the thing the buyer saw and the thing their team connects to are the same object. Never demo one and deliver another. ### Three offers that fit almost every buyer Most MCP engagements fall into one of three shapes. Name them in your proposal and the buyer will pick. | Offer | What it is | Who buys it | How it is priced | | --- | --- | --- | --- | | The public front door | A public server (plus the chat widget) so customers' assistants get correct answers about products, rates, policies and availability | Hotels, clinics, schools, property firms, e-commerce — any business whose website already answers questions | Setup fee plus a monthly retainer for maintenance and sync | | The private brain | A login-only server of SOPs, policies, product facts and internal tools, connected from Claude by staff | Firms of 10 to 500 people with knowledge in too many places: accounting, legal, agencies, operations-heavy SMBs | Per seat, or folded into an existing retainer | | The paid product | A curated server of expertise sold by the seat to outside subscribers | Consultants, training companies, associations, vertical specialists | Per-seat subscriptions with Stripe billing built in; the platform is merchant of record, takes a fee and pays out the rest | Sell the first two as services and the third as a product. The mechanics are identical; the pricing conversation is not. ### How the build works on SuperCognit 1. **Import the source of truth** — Point the platform at the buyer's domain. Pages are crawled into a knowledge base with citations. Upload the PDFs, price lists and internal guides the site does not carry. 2. **Attach the actions** — HTTP tools you configure become callable MCP tools: availability lookups, order status, ticket creation. Add lead capture and every qualified conversation lands in the inbox, a CRM or a webhook. Static "skill" tools return stored content verbatim, so a checklist or a methodology costs nothing per call. 3. **Choose the boundary** — Private (login required, staff only) or public (anyone with an MCP client). One decision per server. Put it on the buyer's own subdomain — mcp.theirbrand.com — before the handover, not after. 4. **Publish both surfaces** — One build gives you a chat agent for the website, WhatsApp and Telegram, and an MCP server for Claude and every other client. Same knowledge, same configuration, answered everywhere. 5. **Set the sync cadence and price the seat** — Frequent for pages that change, slower for the rest. If access is sold, attach a per-seat subscription; checkout, invoices and payouts are handled for you. Plans are priced in EUR with a 7-day trial — long enough to build the first buyer's server before you pitch it. ### Horizontal or vertical: the same server, two ways to sell it A horizontal solution solves one job for every industry: customer support, sales qualification, employee onboarding, policy lookup. A vertical solution solves every job for one industry: the hotel, the clinic, the property agency. MCP servers are unusual in that the same build sells both ways — and the mistake is to pick one when the market wants you to sell the horizontal job in the vertical's vocabulary. #### Selling horizontally: departments Departments buy outcomes they already measure. The knowledge and tools differ; the pitch is the same across industries. | Department | What the server answers | Tools worth attaching | Metric the buyer already tracks | | --- | --- | --- | --- | | Customer support | Products, policies, returns, how-to questions — cited | Order lookup, ticket creation | Tickets deflected, first-response time | | Sales | Pricing, packaging, comparisons, objection answers | Lead capture to CRM, calendar booking | Qualified leads, time to first reply | | Operations | SOPs, checklists, who does what | Static skill tools for procedures | Errors, onboarding time | | HR and onboarding | Policies, benefits, first-week questions | Ticketing for exceptions | Time to productivity, HR inbox volume | | Finance and compliance | Expense rules, approval thresholds, regulatory text | None, deliberately — read-only | Policy exceptions, audit findings | #### Selling vertically: industries Industries buy from people who speak their language. Build the horizontal template once, then sell it in the vertical's words with the vertical's tools. | Industry | The questions it answers all day | The action that makes it a colleague | Public or private | | --- | --- | --- | --- | | Hospitality | Rooms, rates, amenities, check-in, what to do nearby | Availability lookup, booking hand-off | Public front door; private SOP brain for staff | | Real estate | Listings, neighbourhoods, process, fees | Viewing requests into the CRM | Public | | Healthcare and clinics | Services, preparation instructions, insurance, hours | Appointment requests | Public and read-only; strict about what it will not answer | | Education | Programmes, admissions, deadlines, fees | Application follow-up, lead capture | Public | | Professional services | Methodology, engagement scope, compliance rules | Static skill tools, document lookup | Private brain, or paid product for clients | | E-commerce | Products, sizing, shipping, returns | Order status, cart hand-off | Public | | SaaS | Features, pricing, integrations, docs | Support ticket creation, trial sign-up | Public docs server; private for customer-success teams | The productised move is to keep a template per department and clone it per industry: the same support server, dressed in hospitality vocabulary for the hotel and in clinical vocabulary for the clinic. Knowledge and tools change; the structure of the engagement does not, which is what lets you quote it in a day. ### Selling MCP inside a company Not every sale crosses a company boundary. The internal champion — an operations lead, an IT manager, the head of a department — is selling to their own leadership, and the pitch changes shape. Nobody inside cares about the protocol either; they care about the twelve people who ask the same forty questions. The internal case is easier to make than the external one, because the proof is already in the building: the chat thread where the policy was explained for the fourth time, the onboarding doc nobody can find, the spreadsheet of answers that lives on one person's laptop. A private MCP server is that person's knowledge, made available to everyone, inside the assistant they already have open. #### The internal pitch, in three lines - Staff already use Claude and other assistants at work — often without telling anyone. A company server gives them a sanctioned source instead of a guessed one, which is a governance win before it is a productivity win. - It is behind login. Every employee signs in with OAuth; access ends when the person leaves. Nothing is public unless someone makes it so. - It costs a subscription, not a project. No developers, no integration backlog, and it follows the documents as they change. #### Where to start internally One department, one server, the questions that department is tired of answering. Operations SOPs and HR policies are the usual first wins because the content already exists and the audience is the whole company. Publish it, connect it from Claude, and let the head of the department watch their inbox shrink for a month. That number is the budget request for the second server. For an agency, internal servers are also the second sale at every client: the public front door for customers, then the private brain for staff — same workspace, built from the same imports. ### Managing many MCP servers for many customers Once the second customer signs, you are no longer building servers; you are operating a portfolio. The rules that keep a portfolio sane are structural, and they are easier to set up on the first customer than to retrofit on the tenth. #### One workspace per customer, always A workspace is a hard boundary: a server can only answer from what its own workspace holds. Give each customer their own — their knowledge bases, their tools, their credentials, their conversations and leads — and cross-contamination becomes impossible rather than merely unlikely. It also makes offboarding one action: hand the workspace over to the customer or retire it, and no other customer notices. #### One login for your team Under a partner arrangement your workspace is linked to each client workspace, and your staff work inside every one of them from the dashboard login they already have — as administrators, never as owners. Access is derived from the partnership rather than from membership rows, so a client cannot accidentally remove you mid-engagement, and ending the relationship is one change on your side. When a client wants to take a server in-house, the workspace is already theirs; you step out rather than migrate. #### The operating rhythm 1. **Name things like an operator** — Workspaces named for the client, servers named for the audience ("Acme — customer front door", "Acme — staff SOPs"), hostnames on the client's domain. Six months from now someone will need to find the right one in a hurry. 2. **Standardise the build, not the content** — A template per offer: which knowledge bases, which tools, which sync cadence, public or private. New client, clone the template, swap the sources. You can quote it in a day because you have built it twenty times. 3. **Set sync cadence per source** — Rates and availability hourly; policies weekly; the about page whenever. Maintenance is then true without being expensive, which is what makes the retainer honest. 4. **Watch the conversations** — Every server's transcripts and captured leads sit in the workspace inbox. A weekly pass over "questions we could not answer" is your content roadmap and your upsell list in one place. 5. **Bill the way the client wants** — Consolidated wholesale seats invoiced by you, or the client paying the platform directly with your margin applied. Per seat either way; nobody forecasts tokens. 6. **Automate provisioning from Claude** — The platform has its own MCP server. Connect it from Claude and create agents, MCP servers, knowledge bases and tools conversationally — or script the onboarding of the next twenty clients from a checklist. The operator of many servers should not be clicking through a dashboard twenty times. #### What isolation buys you commercially Isolation is not only a security property; it is the reason you can say yes to competitors. Two hotels on the same street, two accounting firms in the same city: separate workspaces, separate servers, separate domains, and no prompt engineering doing the separating. Say it in the proposal. The buyers who ask about it are the ones worth having. ### Pricing that survives the second conversation Three anchors keep pricing simple across all three offers. - Price against the alternative. A seat that saves one email to a person a month is worth a fraction of that person's hour, and the buyer already knows what their people cost. - Keep the menu short. Setup fee plus retainer for services; one per-seat price for products. A buyer forced to compare tiers ends up comparing you to doing nothing. - The test is approval, not accuracy. If the buyer can say yes without calling a meeting, the price is right. For products sold by the seat, the platform handles checkout, recurring invoices and payouts as merchant of record and takes a platform fee; your accounting sees payouts. For agency retainers, bill as you already do and treat the seat cost as a line item. For partners with many clients, wholesale seat pricing under an agreement is the same conversation with a smaller number in it. ### What not to promise The buyers who renew are the ones you were straight with. Four lines to keep out of every proposal: - It does not replace judgment. The server handles the repeatable layer; the bespoke decision still needs a person. Subscribers who outgrow the answers are the consulting pipeline, not a failure. - It does not know what was never published. A server is as good as its sources, and "we'll add that document" is a legitimate answer to a wrong demo. - Public means public. A front-door server is a deliberate decision about what a stranger's assistant may ask. Curate it; do not let it accumulate. - Cheap tokens are not free tokens. Answering costs something per question; static tools cost nothing per call, which is why methodology products should lean on them. ### A 30-day plan for the first three customers 1. **Week one: your own server** — Import your own site and documents, connect it from Claude, and live with it. You cannot sell what you have not used, and your objections are the buyer's objections. 2. **Week two: three built demos** — Pick three prospects whose websites already answer questions. Import each, run the three-questions test yourself, and fix the gaps before anyone else sees them. 3. **Week three: three meetings** — Arrive with the server built. Let them ask it things. Send the proposal the same afternoon with one of the three offers named. 4. **Week four: operate** — Workspaces named, hostnames on their domains, sync cadences set, inbox reviewed weekly. Then start the next three. Selling MCP is selling the end of a specific, familiar frustration: the assistant that guesses. The protocol is what makes the fix portable; the platform is what makes it deliverable without code; the offer is what makes it a business. Build one this week and sell it to someone who already trusts you. ## How to automate WhatsApp Business without a developer (step by step) https://supercognit.com/blog/how-to-automate-whatsapp-business · 2026-09-10 You can automate WhatsApp Business in an afternoon: connect your number to an AI agent built from your own website, and it answers, qualifies and hands off around the clock. Here is the setup step by step, what to automate first, what to keep human, and what it costs once Meta bills every reply from 1 October 2026. To automate WhatsApp Business you need three things: a WhatsApp Business number on Meta's Cloud API, an agent that actually knows your business, and a place where the conversations land so a person can step in. On SuperCognit that is one afternoon of work and no code — import your website, connect the number from the dashboard, publish. This guide walks through it in order, then covers what to automate first, what to keep human, and how the economics changed now that Meta charges for every reply. ### What "automation" should mean on WhatsApp Most WhatsApp automation is a keyword menu: press 1 for opening hours, press 2 for prices. Customers do not write like that. They write full sentences, in their own language, at eleven at night, and they expect the answer a good employee would give. Automation worth having looks like this: - An instant first reply that answers the actual question, grounded in your website and documents and citing the page it came from. - Follow-up questions that qualify — dates, party size, budget, what the customer needs — asked in conversation, not as a form. - A clean handoff: when the agent cannot help, the thread is in your inbox with the whole context, not a lost message. - Lead capture with consent, so a warm enquiry becomes a contact in your CRM while the customer is still typing. - Payment links in the thread where a deposit or an order is the natural next step. ### Before you start: three prerequisites - A WhatsApp Business number on the Cloud API. That means a Meta business portfolio and a number that is not currently registered in the consumer or small-business WhatsApp app — a number can only live in one place. - A website that is roughly current. It becomes the agent's knowledge, so the rate sheet on your site is the rate sheet the agent quotes. Fix the obviously stale pages first; the sync will keep them current afterwards. - A decision about who watches the inbox. Automation handles the volume; someone still owns the exceptions. ### The setup, step by step 1. **Build the agent from your website** — Point SuperCognit at your domain. Pages are crawled into a cited knowledge base and a first agent is drafted in about two minutes. Upload the documents the site does not carry — price lists, policies, the FAQ that lives in someone's sent folder. 2. **Set the tone and the rules** — Write the welcome message, tell the agent what it must never do (quote discounts it cannot cite, give medical or legal advice, promise stock it has not checked), and add the one-line disclosure that it is an AI — the EU AI Act requires it at the start of the conversation. Language is automatic: the agent replies in whatever language the customer writes, across 12 and more. 3. **Connect the number** — In the dashboard, open the agent's channels and add WhatsApp. Where Meta's sign-in popup is available you authorise the number in a few clicks; otherwise paste the Cloud API details from your Meta app — phone number ID, a permanent access token and the app secret, all stored encrypted. The dashboard then shows the callback URL and verify token to paste into the Meta app's WhatsApp configuration, and you subscribe to the messages field. No code at any point. 4. **Send yourself the first twenty messages** — Ask the questions your customers really ask, in the way they ask them, including the badly spelled ones. Read the answers and the citations. Every wrong or missing answer is a page to fix or a document to upload — this half hour is where the quality comes from. 5. **Turn on lead capture and handoff** — Tell the agent, in plain language, what makes an enquiry worth a human: the use case, the dates, the size of the order. Connect Pipedrive or HubSpot with one click, or let leads land in the platform inbox with a webhook to wherever you want them. Consent is asked before contact details are stored. 6. **Go live and set the sync** — Publish. Then set the re-sync cadence per source: hourly for the pages that change (availability, menus, prices), weekly for the rest. From here the agent tracks your site without a ticket. ### What to automate first, and what to keep human | Conversation type | Automate? | Why | | --- | --- | --- | | Opening hours, location, prices, policies | Yes, from day one | Answered from cited pages; the highest volume and the lowest risk | | "Is it available on these dates?" | Yes, if you have an API | An HTTP tool checks your booking engine; without one, the agent qualifies and hands over the booking link | | Order and delivery status | Yes, with a tool | Look up the order by number instead of describing how one would | | Lead qualification | Yes | Consent, qualifying questions, CRM write with the transcript attached | | Complaints and refunds | Keep human | The agent collects the facts and hands off; the decision is a person's | | Anything involving money | Partly | The agent sends a Stripe checkout link; it never negotiates or takes card details in chat | ### The economics from 1 October 2026 From 1 October 2026, Meta charges for service messages — the free-form replies inside the 24-hour customer window that were free until then. An automated conversation that exchanges twenty messages is twenty billed messages. Three habits keep the bill sane: make each answer complete so the thread is short; move volume to channels you own — the widget on your site and Telegram carry no per-message fee — and keep WhatsApp for the customers who prefer it; and use daily spend caps and session limits so a runaway thread cannot surprise you. The point is not to leave WhatsApp. It is to stop it being the only door. ### Compliance, in two lines Say it is an AI in the first message, in the customer's language. Ask before you store contact details, and keep the transcripts — they are your record of what was said, to whom, and when. Both come out of the box; both are still your responsibility. ### If you run WhatsApp for clients Agencies run the same playbook per client with one structural rule: one workspace per client, so knowledge, numbers, credentials and conversations never mix. Your team works across every client workspace from one login under a partner arrangement, the client keeps ownership, and you bill it as a retainer or per seat. Build the first client's agent before the pitch; the WhatsApp demo on their own number is what closes it. ### What you should have by the end of the afternoon A number that answers in seconds, in the customer's language, from your own content, with every thread visible to your team and every warm enquiry in your CRM. Plans are priced in EUR with a 7-day trial — long enough to connect the number, send yourself the twenty messages, and see whether the answers are good enough to leave running overnight. ## AI reservation agent for hotels: how it checks availability, quotes and books https://supercognit.com/blog/ai-reservation-agent-hotels · 2026-09-10 A hotel AI reservation agent answers the questions guests ask before they book, checks availability through your booking engine, and either completes the handoff or captures the enquiry — around the clock, in the guest's language. Here is how it works, what it connects to, and how group and event bookings fit in. An AI reservation agent for a hotel is an agent that handles the conversation before the booking: it answers the questions guests ask first — what is in the room, is the pool heated, how far is the beach — checks availability through your booking engine or PMS, and then either sends the guest to the right booking step or captures the enquiry for your reservations team. It works on your website, on WhatsApp and on a branded page, in the guest's own language, at three in the morning. On SuperCognit it is built from your website without code and connected to your booking system through HTTP tools. This is how the pieces fit. ### What guests actually ask before they book Reservations teams know this list by heart, because they answer it hundreds of times a season. Four Seasons Fairways, a resort in the Algarve running its concierge on SuperCognit, sees the same questions every day across 132 villas and apartments: - Availability for specific dates, and what changes if the dates move by a day. - What is actually in a Deluxe Villa, and how it differs from the next category up. - Whether the pool is heated, whether there is a kids club, whether the restaurant needs a reservation. - Transfers from the airport, parking, late check-out, early check-in. - The cancellation policy, in plain words, before anyone commits. None of these is a booking. All of them come before one — and a guest who cannot get the answer at 23:00 books somewhere that answers. ### The three layers of a reservation agent #### Knowledge: your website, cited The agent is built from your own pages — rooms, rates, amenities, policies, restaurants — crawled into a knowledge base that cites its source on every answer. Seasonal pages re-sync on a schedule, as often as hourly, so the offer on your site is the offer the agent describes. It answers in the guest's language automatically, across 12 and more. #### Actions: your booking engine, through tools SuperCognit deliberately does not ship its own calendar. Instead the agent calls the system you already run: an HTTP tool wired to your booking engine, PMS or channel manager lets it check availability for the guest's dates and create an enquiry or a hold, where the API allows it. For systems without an API, the agent still qualifies the request and hands the guest the right booking link at the right moment — not a homepage and good luck. Where a deposit is part of the flow, it can send a Stripe checkout link inside the conversation. #### Handoff: your reservations team, with context Every conversation lands in the workspace inbox with the guest's questions and the agent's answers. Enquiries the agent could not complete — and any it captured as a lead — reach your team or your CRM with the transcript attached, so the first human reply starts informed instead of asking the guest to repeat themselves. ### A booking conversation, step by step 1. **The guest asks a real question** — "Do you have a two-bedroom villa for the second week of June, and is it close to the pool?" The agent answers the second half from your pages, with the citation, before touching availability. 2. **It qualifies what it needs** — Dates, number of guests, children's ages, preferences the guest mentioned. Collected in conversation, so what reaches the booking engine is a complete request. 3. **It checks availability** — Through the HTTP tool, against your live system. It presents what is actually free at the actual rate — nothing it did not verify. 4. **It completes or hands off** — If the guest is ready, it sends them to the booking page for that category and those dates, or creates the enquiry in your system. If a deposit is needed, the checkout link goes in the thread. 5. **It catches the not-yet-ready** — "Let me ask my partner" is the most common ending. The agent offers to save the enquiry, asks for consent, and the guest becomes a lead your team can follow up — instead of a closed tab. ### What it must never do - Confirm availability it did not check. Without a tool, it qualifies and hands over; it does not guess. - Invent rates, discounts or upgrades. Every number comes from a page or an API response, and it says so. - Take card details in the chat. Payment is a link to Stripe checkout, never a conversation. - Handle a complaint as if it were a booking. It collects the facts and gets a person. ### Group and event bookings Weddings, corporate groups, tour operators and family reunions do not book through a calendar, and they should not be automated end to end. What an agent does well here is the intake: it qualifies dates and flexibility, group size and rooms needed, the type of event, catering and space requirements, and a budget window — then captures the whole request, with consent, as a lead in your CRM or inbox with the transcript attached. Your events team opens a complete brief instead of a two-line email, and the quote is still a human's. The value is not replacing the sales conversation; it is making sure every serious enquiry arrives whole, including the ones that came in on a Sunday. ### Which systems it connects to | Your situation | What the agent does | | --- | --- | | Booking engine or PMS with an HTTP API | Checks availability and creates enquiries or holds through a tool you configure in the dashboard | | Channel manager or engine with a booking URL but no API | Qualifies the request and sends the guest to the right booking page with the dates prefilled where the URL allows | | No online booking at all | Captures the enquiry as a lead with consent; your team confirms by email or phone | | Deposits or prepayment | Sends a Stripe checkout link and receipt inside the conversation | ### Where it runs The same agent runs as a widget on your website, on a branded page such as concierge.yourhotel.com with the certificate handled for you, and on WhatsApp and Telegram. One knowledge base, one set of rules, one inbox — a guest who asks on WhatsApp gets the same answer as one who asks on the site, because there is one agent, not one per channel. ### Setting it up 1. **Import the website** — Rooms, rates, amenities and policies become a cited knowledge base in about two minutes. Upload the documents the site does not carry: seasonal menus, the group brochure, the cancellation terms in full. 2. **Wire the booking engine** — Add an HTTP tool for availability and, if the API supports it, for creating enquiries. Describe each tool in plain words; the description is what the agent reads to decide when to use it. 3. **Set the boundaries** — What it may quote, what it must hand off, the disclosure line that it is an AI. Then send it the twenty questions your front desk answers most and read every answer. 4. **Publish everywhere and watch the inbox** — Widget, branded page, WhatsApp. Review the first week's conversations: the questions it could not answer are the pages you need to write. Plans are priced in EUR with a 7-day trial. Import your site today, ask the agent your three most common pre-booking questions tonight, and you will know by tomorrow whether it belongs on your website. ## Wonderchat vs Intercom Fin for lead qualification: which one actually qualifies? https://supercognit.com/blog/wonderchat-vs-intercom-fin-lead-qualification · 2026-09-10 Wonderchat is a fast website chatbot; Intercom Fin is a support agent inside Intercom. Neither was built to qualify leads and write them into a CRM. Here is how each handles lead qualification, a side-by-side table, and a third option to test against both. Short answer: Wonderchat and Intercom Fin are built for different jobs, and lead qualification is not the centre of either. Wonderchat gets a chatbot live from your website in minutes and answers well; its lead handling is light. Intercom Fin is a strong support agent, but it lives inside Intercom, is priced per resolved conversation, and is designed to close tickets rather than open deals. If qualifying leads is the job, evaluate a third shape alongside them: an agent that asks your qualifying questions and writes contacts and deals into your CRM with the transcript attached. This comparison is based on the vendors' public materials at the time of writing — plans change, so check the current ones before you decide. ### What lead qualification has to include Before comparing tools, agree on the job. A chatbot qualifies leads when it does all of the following, not one of them: - Answers the visitor's product questions first, from your own content, so the conversation is worth having. - Asks the qualifying questions you defined — use case, timeline, size, budget — in conversation, not as a form. - Asks for consent before storing contact details. - Writes a contact and a deal into your CRM, in the pipeline and stage you chose, with the whole transcript attached. - Works on WhatsApp and other channels with the same rules, not only in the website bubble. - Gives unqualified visitors real answers without creating a task for a rep. ### Wonderchat for lead qualification Wonderchat's strength is the start: point it at your site and documents and you have a chatbot embedded in minutes, answering with citations. That is genuinely useful, and for a small site it may be all you need. For qualification the limits show quickly. Its public materials describe it as answer-focused, with basic handoff and limited channels beyond the website widget. Collecting a visitor's contact details is a different thing from asking three qualifying questions and creating a deal in HubSpot in the right stage — test that specific path before assuming it exists. ### Intercom Fin for lead qualification Fin is a serious support agent, and its inbox and human handoff are core strengths — if your team already runs on Intercom. Two things matter for lead qualification. First, it is designed around resolving support conversations inside Intercom's platform, so sales qualification is a secondary use of a support tool. Second, it is priced per resolution on top of Intercom seats, which means the bill grows exactly when the agent is busy. For a support team that already pays for Intercom that may be fine; for a sales-led site evaluating a qualification bot, it is a lot of platform for the job. ### Side by side | Criterion | Wonderchat | Intercom Fin | SuperCognit | | --- | --- | --- | --- | | Primary job | Website chatbot from your site | Support automation inside Intercom | Build and run AI agents that answer, qualify and act | | Time to a live bot | Minutes | Depends on Intercom setup | Minutes — website import drafts the agent | | Answers with citations | Yes | From your help centre | Yes — hybrid retrieval | | Asks configurable qualifying questions | Check current plans | Support-task focused | Yes — defined in plain language | | Writes deals to HubSpot or Pipedrive | Check current plans | Via Intercom's ecosystem | Native, one-click OAuth, transcript attached | | WhatsApp and Telegram | Limited | WhatsApp via Intercom | Native | | Pricing shape | Message quotas per plan | Per resolution plus seats | Plans with metered usage and hard cost caps | | Own domain | Embed and share links | Messenger widget | Yes, automated SSL | | Sell access to the agent | No | No | Built in | Product names belong to their respective owners. SuperCognit is not affiliated with Wonderchat or Intercom, and the rows above summarise public materials rather than a formal test of either product. ### The e-commerce question A recurring version of this search is "does it qualify leads for e-commerce?" — and for a shop, qualification usually means something narrower than for a B2B site. The visitor is not being scored on budget and timeline; they want sizing, shipping and returns answered, an order looked up, and a nudge to the right product or cart. That makes the answering layer table stakes and the actions the differentiator: an order-status lookup through your store's API, a hand-off to the cart, a checkout link in the thread. Judge any of the three tools on those actions, not on whether the bubble can collect an email. ### How to test all three in a week 1. **Import your site into each** — The same pages, the same documents. If a tool cannot start from your website, that is your first data point. 2. **Write your three qualifying questions** — The ones your best rep asks in the first two minutes. Configure them, in whatever way each tool allows. 3. **Run twenty real conversations** — Use the questions your visitors actually ask, including the ones that should not become leads. 4. **Open the CRM** — What landed? A contact with a transcript in the right stage, a bare email, or nothing? This is the whole comparison in one screen. 5. **Check the unqualified path** — The visitor who was just browsing should have got good answers and created no task for anyone. ### Verdict If you want a fast website chatbot and leads are a nice-to-have, Wonderchat is a reasonable pick. If your company already runs on Intercom and support deflection is the goal, Fin is the natural fit. If lead qualification is the actual job — questions asked, consent taken, deals written, WhatsApp included — test an agent built for that against both. SuperCognit's plans are priced in EUR with a 7-day trial, which is enough time to run the twenty conversations and look at the CRM. ## How to send chatbot conversations and leads to a webhook (n8n, Make, Zapier) https://supercognit.com/blog/chatbot-webhook-n8n-make-zapier · 2026-09-10 A chatbot webhook is an HTTPS endpoint your chatbot platform POSTs JSON to when something happens — a new conversation, a finished reply, a captured lead. Here are the events, the payloads, how to receive them in n8n, Make and Zapier, how to verify the signature, and the mistakes that lose data. A chatbot webhook is an HTTPS endpoint that your chatbot platform POSTs a JSON payload to whenever something happens: a visitor starts a conversation, the agent finishes a reply, a lead is captured. Point that endpoint at n8n, Make or Zapier and everything downstream — Slack alerts, spreadsheets, CRM updates, follow-up emails — is automation you already know how to build. On SuperCognit you register the endpoint from the dashboard, choose the events, and every delivery is signed and retried. This guide covers the events that exist, what the payloads look like, the receiver in each tool, and the handful of mistakes that quietly lose data. ### The events you can subscribe to Event names are a public contract: they are added to, never renamed, so an automation built today keeps working. The ones most people use: | Event | Fires when | Typical use | | --- | --- | --- | | conversation.created | A visitor starts a new conversation with an agent | Count conversations by page or channel; start a session timer | | message.created | A message is added, by either side | Full transcript mirroring into your own store | | message.completed | The agent finishes generating a reply | Quality review, sentiment scoring, token accounting | | lead.captured | The agent captured a qualified lead | Slack alert, spreadsheet row, CRM you run yourself | | lead.delivered | The lead reached the connected CRM | Confirm the handoff; fires later than captured, after any retry | | lead.failed | Delivery to the CRM was given up | Alert a human — the lead is still in the dashboard and re-sendable | | payment.completed | A checkout link sent in chat was paid | Fulfilment, receipts, revenue dashboards | | knowledge.source.failed | A website or document re-sync failed | Alert whoever owns the content | | product.published / unpublished | An agent or MCP server changed status | Change log, status page | | channel.disconnected | WhatsApp or Telegram lost its connection | Page the on-call person before customers notice | An endpoint with an empty event list receives everything. Start narrow — one event, one workflow — and widen once the first recipe is boring. ### What a lead payload looks like ``` { "id": "evt_…", // stable across retries — dedupe on this "event": "lead.captured", "version": 1, "createdAt": "2026-09-10T09:14:02.000Z", "workspaceId": "…", // a receiver may serve several workspaces "data": { "leadId": "…", "productId": "…", "sessionId": "…", "agent": "Showroom assistant", "contact": { "name": "Ana Silva", "email": "ana@example.com", "phone": null }, "fields": { "budget": "20k" }, "consent": true, "sourceUrl": "https://acme.com/pricing", "status": "pending", "crm": { "provider": "pipedrive", "leadId": "…", "url": "…" } } } ``` The fields object carries whatever qualifying answers the agent collected, consent records that the visitor agreed to be contacted, and crm is present only when a CRM is connected. Payloads only ever gain fields; nothing you map today will disappear. ### Register the endpoint 1. **Create the receiver first** — In n8n, Make or Zapier, add a webhook trigger and copy the URL it gives you. It has to accept POST and answer with a 2xx quickly — do the slow work after responding. 2. **Add the webhook in the dashboard** — Under Settings → Webhooks, add an endpoint, paste the URL and pick the events. Copy the signing secret; you will not build the verification without it. The same registration is available over the REST API for teams that provision from code. 3. **Send a test** — Start a conversation with your own agent, or capture a test lead. Watch the delivery log in the dashboard: each attempt is recorded with the response status your endpoint returned. 4. **Verify the signature before trusting the body** — Every delivery carries an x-supercognit-signature header: sha256= followed by the HMAC-SHA256 of the raw request body, keyed with your secret. Compute it, compare in constant time, reject on mismatch. Anyone can POST to a public URL; only the platform can sign. ``` // Node.js — verify the signature on the RAW body (not the parsed JSON) import { createHmac, timingSafeEqual } from "node:crypto"; export function verify(rawBody, header, secret) { const expected = "sha256=" + createHmac("sha256", secret).update(rawBody).digest("hex"); return expected.length === header.length && timingSafeEqual(Buffer.from(expected), Buffer.from(header)); } ``` ### Receiving in each tool #### n8n Use the Webhook node with method POST and "Respond immediately". Add a Code node right after it to verify the signature with the snippet above — n8n exposes the raw body when you enable it on the node — then branch on the event field with a Switch node. Self-hosted n8n behind your own domain is the tidiest option: your secret and your data stay with you. #### Make Create a Custom webhook, run one test delivery so Make learns the structure, then add a router keyed on event. Make can compute an HMAC in a text function; compare it to the header before the router, and stop the scenario on mismatch. #### Zapier Use Webhooks by Zapier with Catch Hook as the trigger. Zapier parses the JSON for you, which is convenient and also means you need a Code step to reconstruct the raw body for verification; if that is more than you want, restrict the Zap to low-risk recipes such as notifications and keep CRM writes on the native integration. ### Retries and idempotency Deliveries that do not get a 2xx are retried up to five attempts with exponential backoff — one, two, four and eight minutes apart — and every attempt is logged with the response status. The id on the envelope stays the same across all attempts, which is what lets you dedupe: if your endpoint answered slowly and the retry arrived anyway, the second copy has the same id, and you skip it. Answer fast, store the id, do the work afterwards. ### Five recipes that pay for the setup - lead.captured → Slack message to the sales channel with the contact, the qualifying answers and a link to the transcript. - lead.failed → alert to whoever owns the CRM connection, with the lead id to re-send from the dashboard. - message.completed → append to a review sheet, then sample ten a week for quality. - conversation.created → count by sourceUrl in a dashboard, so you see which pages start conversations. - knowledge.source.failed → ticket to the content owner before the agent starts answering from stale pages. ### "ChatGPT webhook": why that search finds nothing useful People searching for a ChatGPT webhook usually mean one of two things. If they built a custom GPT, the answer is that a GPT can call your API through an action, but it does not push you an event when a conversation happens — there is no outbound webhook to subscribe to. If they built a bot on the OpenAI API themselves, the webhook is their own code to write. Either way the feature being looked for is a platform that emits events about conversations and leads. That is what the event catalogue above is. ### Mistakes that lose data - Doing the slow work before responding. A CRM write that takes twelve seconds times out the delivery, which then retries — and now you have two deals. - Verifying the parsed JSON instead of the raw body. Re-serialising changes bytes; the signature will never match. - Subscribing to everything on day one. message.created alone can be thousands of events a day; start with the one you will act on. - Ignoring lead.failed. The lead is safe in the dashboard, but nobody knows to re-send it unless something tells them. Webhooks are the part of a chatbot nobody sees and everybody depends on. Register one endpoint, verify one signature, build one recipe, and the rest of your stack starts hearing what your agent is doing. ## MCP was never just for software — Anthropic just proved it https://supercognit.com/blog/mcp-model-hardware-standard · 2026-08-31 On 28 August Anthropic opened a research preview letting AI agents operate lab and manufacturing hardware through MCP-compatible interfaces. It has nothing to do with a support chatbot — and it is still worth twenty minutes, because it tests whether the protocol we built on actually generalizes. Every post we have written about MCP so far has been about software: tools, knowledge bases, OAuth, session handling. On 28 August, Anthropic opened a research preview of something in a different category entirely — the Model Hardware Standard, a specification for letting AI agents discover, operate and troubleshoot physical equipment. Microscopes, liquid handlers, robotic arms. ### What the standard actually does The Model Hardware Standard gives a device a common driver interface and a natural-language description an agent can read, the same idea an MCP tool description already uses for an API. Anthropic says it works with any device that exposes a programmable control surface, is model-agnostic, and is built to be reachable through standard protocols — MCP among them — so an agent harness does not need a bespoke integration per instrument. ### Where it came from and who is testing it - Development started as a collaboration between Anthropic and the HHMI Janelia Research Campus. - The research preview is limited for now to a first group of scientific research labs and advanced manufacturers, not a general release. - Anthropic says early use includes automating a protein-assay procedure at Genentech and running drug-discovery experiments at Carnegie Mellon University. - Doosan Robotics, Tecan, Universal Robots, Hugging Face and Raspberry Pi are named as partners building or testing integrations. - The pitch is integration time cut from weeks or months down to hours — the same argument made for MCP itself when it replaced bespoke API integrations with one description an agent can read. ### Why this belongs in an MCP roundup and not a robotics one We wrote in June about MCP's move to the Agentic AI Foundation, and the argument then was that governance is what makes a protocol a safe bet: not owned by one vendor, moving through a public process, a real deprecation window. This is a different kind of evidence for the same bet. A protocol designed for one thing — an agent calling a software tool — turning out to describe a robotic arm with the same shape of interface is a sign the abstraction was right, not just convenient. ### What this does not mean for a business chatbot Nothing about the Model Hardware Standard is relevant to a support agent, a lead-qualification flow or a knowledge base — it is a research preview scoped to labs and factory floors, and SuperCognit's own MCP servers stay exactly what they are: HTTP tools, knowledge, and the workspace resources a business actually needs an agent to reach. We are not adding a hardware angle because one exists in a lab somewhere. The reason to notice this at all is narrower and more useful: the same open interface layer we built our MCP product on keeps getting reused for things nobody designed it for on day one, which is a better signal than any single feature release. ## The MCP checklist: what a professional company MCP needs before you share it https://supercognit.com/blog/mcp-checklist-professional · 2026-08-30 A professional company MCP server needs six things before you share it: current-spec auth, a clear data boundary, cited answers, fresh content, accountable tools and a domain that says who it speaks for. The full MCP checklist, with checks you can run. A professional company MCP server needs six things before you share it: authentication that matches the current spec, a data boundary you can explain in one sentence, answers that cite their sources, content as fresh as the business, tools you can account for, and a name that says who it speaks for. Everything below is that list, expanded into checks you can actually run. ### 1. Auth that matches the current spec MCP auth has moved fast, and clients follow the spec, not your release schedule. Current means OAuth 2.1: users sign in as themselves, tokens carry an issuer the server validates, and client registration follows the CIMD flow so new MCP clients can connect without a manual setup ticket. SuperCognit servers implement this out of the box; if you built yours by hand against an older draft, fix this first, because it is the thing that fails at connection time — in front of the person you shared it with. - Every user authenticates individually — no shared key in a config file that outlives the person who pasted it. - Issuer validation is on. A token from the wrong place is a no. - Connecting works from Claude and from at least one other MCP client. Test both before anyone else does. ### 2. A boundary you can explain in one sentence The sentence is: this server answers from this workspace, and nothing else. Per-workspace isolation makes that structural rather than aspirational — the server cannot reach knowledge that is not in its workspace. Your job is the deliberate part: deciding what goes in, and whether the server is private or public. A private server is an internal company brain — SOPs, policies, product facts for your own team. A public one speaks for the company to anyone who connects. Those are different products, and the choice should have a name attached to it, not a default. ### 3. Answers that show their sources A company MCP gets quoted — in documents, in decisions, eventually in a meeting where someone asks where that came from. Cited answers make that moment boring instead of career-limiting. When the knowledge base is built by crawling your site, every answer points back to the page it came from; uploaded documents are cited the same way. - Ask the ten questions your team actually asks, and follow every citation back to its source. - The same build produces a chat agent — use it as the smoke test. If the chat answers are wrong, the MCP answers are wrong with extra steps. - Nothing in the knowledge base would embarrass you quoted verbatim. It will be quoted verbatim. ### 4. Content as fresh as the business Stale answers are worse than no answers, because they arrive with confidence. Knowledge re-syncs on a schedule — as often as hourly — and the cadence is a decision per source, not a global toggle to set and forget. Pricing, availability and policy pages earn a fast schedule; the company history page can wait. For uploaded documents the sync cannot save you: every document needs an owner who replaces it when reality changes, or the server becomes a very confident archive. ### 5. Tools you can account for Knowledge answers questions; tools take actions. HTTP tools you attach become callable MCP tools, which means anyone connected can invoke them — so the standard is stricter than it is for content. Before sharing, you should be able to say, for every tool on the server: what it can touch, what the worst plausible call does, and why it is attached at all. A short tool list is a feature, not a gap. A server is judged by what it can do, not only by what it knows. ### 6. A name that says who it speaks for Custom domains are supported, and for a company server they are not cosmetic. A team connecting to your own hostname knows whose product this is and who stands behind its answers; a platform URL says the opposite. If the server is going to speak for the company, it should live at the company's name. Notice the pattern in the list: none of it is code, all of it is decisions — what goes in, who connects, how fresh, what it can do, whose name is on it. The build itself is the fast part; a crawled, cited knowledge base exists about two minutes after you import the site. Spend the time you saved on the checklist. ## How to sell access to your MCP server: pricing, seats and Stripe payouts https://supercognit.com/blog/sell-access-to-your-mcp-server · 2026-08-26 To sell access to your MCP server you need to solve three things: who may connect, what a seat costs, and how the money reaches you. Here is how per-seat pricing and built-in Stripe subscriptions and payouts work in practice. There are three problems to solve when you sell access to an MCP server: who may connect, what exactly a customer is buying, and how the money reaches your account. On SuperCognit all three are handled — servers authenticate every user with OAuth 2.1, access is sold per seat, and Stripe subscriptions, invoices and payouts run through the platform, with SuperCognit as merchant of record taking a platform fee. What is left to you is the product and the price. This post is about those, and about what the built-in parts actually do. ### Why the seat is the right unit MCP access is personal by construction. Every user who connects your server from Claude or any other MCP client authenticates through OAuth 2.1 — issuer validation and CIMD registration per the current spec — so the platform always knows which human is asking. A seat is simply that authenticated person, made billable. Seats are also the unit buyers already understand. A company does not want to forecast query volumes for a knowledge product; it wants to know what ten people cost per month, approve it once, and stop thinking about it. Per-seat pricing is legible on both sides of the invoice, which is most of what a pricing model has to achieve. ### Pricing without a spreadsheet There is no formula, but three anchors keep you out of the weeds: - Price against the alternative, not the cost. If a seat saves someone one email to you per month, the seat is worth a fraction of what that email costs in your time — and your time is the expensive line item. - Keep tiers few. One per-seat price beats a matrix; a buyer forced to compare plans will end up comparing you to doing nothing. - The test is not whether the price is right. It is whether the buyer can approve it without calling a meeting. ### What the subscription is buying A fair question from buyers: why is this recurring? Because the product is not a file, it is a maintained answer surface. The knowledge base re-syncs on a schedule — as often as hourly — so it tracks your source material; answers stay cited so the buyer's team can verify them; and the server keeps up with the MCP spec, whose auth has already changed once and will change again. A PDF is a one-time sale because it starts rotting on delivery. This does not. ### From published server to first payout 1. **Publish the product** — Import your site — crawled into a cited knowledge base in about two minutes — upload your documents, attach any tools. The same build gives you a chat agent, which is the demo you will send prospects before they commit to a seat. 2. **Set the per-seat price** — Attach a per-seat subscription to the server. Checkout, recurring billing and invoices are handled by the platform; you write no billing code and operate no payment stack. 3. **Put it on your domain** — Custom domains are supported. Buyers should connect to your hostname — it is your product, and the address should say so. 4. **The buyer subscribes and seats their team** — Each person connects from Claude or any MCP client and signs in through OAuth 2.1. Access follows the seat — not a shared key that outlives the employee who pasted it into a config file. 5. **The payout arrives** — SuperCognit is the merchant of record: the subscription, the invoice and the payment run through the platform, the platform fee comes off, and the remainder is paid out to you. ### What merchant of record means in practice The merchant of record is the party actually making the sale — the name on the transaction, the issuer of the invoice. Here that is SuperCognit, which is precisely the point: for a consultant or a small firm, the alternative to a merchant-of-record arrangement is becoming a part-time billing department. The platform fee is what that department costs. Your accounting sees payouts, not payment operations. ### Keep the paid boundary honest The server answers only from its own workspace — isolation is structural — so keep the paid knowledge base deliberate. You can run a private internal server and a paid public one side by side without either seeing the other; what you should not do is let the paid product accumulate content nobody decided to sell. Buyers are paying for a curated surface. Curate it. None of this requires the product to be finished. It requires the product to be honest about what it knows. Publish, price a seat, sell it to someone who already trusts you, and let the invoice history tell you when to raise the price. ## AI roleplay: let reps practice objections against your own playbook https://supercognit.com/blog/ai-sales-roleplay-training · 2026-08-25 Sales training fails in the gap between the workshop and the next real call. An AI roleplay agent built from your own playbook gives reps unlimited practice reps — here's how to set one up in an afternoon. Every sales leader has watched it happen: the workshop lands, everyone nods, and three weeks later the calls sound exactly like they did before. The problem isn't the training content — it's that reps get almost no repetitions between the workshop and the next live prospect, because practice requires a partner and managers don't scale. An AI roleplay agent closes that gap. Not a generic chatbot playing 'difficult customer' — an agent built from your playbook, your objection list and your pricing rules, prompted to behave like the buyer you actually sell to. ### What a playbook-grounded roleplay agent does - Plays the persona you configure — the sceptical CFO, the technical evaluator, the champion with no budget — and pushes back with the objections from your own list, not invented ones. - Lets a rep run the same objection ten times in a row, out loud or in writing, at 11pm before the big call, without booking anyone's calendar. - Answers consistently from approved content, so practice reinforces the current playbook rather than a colleague's folklore version of it. - Leaves a transcript, so enablement can see which objections the team keeps fumbling and coach on real material. ### Setting it up on SuperCognit 1. **Load the playbook** — Upload your playbook, battle cards and pricing one-pagers as knowledge, or import the pages from your wiki. The agent answers with citations, so reps can jump to the source. 2. **Write the persona prompt** — Describe the buyer: role, incentives, the objections they raise and how hard they push. The visual prompt editor keeps this readable enough that a sales manager — not an engineer — owns it. 3. **Gate it to your team** — Set the agent to login-required so it stays internal. Reps sign in and practice; nothing leaks to your website. 4. **Review transcripts weekly** — The Inbox shows every practice session. Ten minutes a week tells you which objection is costing deals — that becomes the next team session's agenda. ### What this doesn't replace A roleplay agent is not a sales methodology, and it won't hold anyone accountable for pipeline hygiene — tools built for enforcement do that. What it does is make the practice layer free and unlimited, which is the part human coaching never scales to. Trainers set the standard; the agent supplies the repetitions. A workspace on the Starter plan is enough to run a roleplay agent, a rep-facing knowledge assistant and a public qualification agent side by side — they can even share the same playbook knowledge. ## AI lead capture and qualification chatbot: from first message to CRM and booked meeting https://supercognit.com/blog/ai-lead-qualification-crm · 2026-08-25 A lead capture chatbot answers the visitor's questions, asks the qualifying ones, takes consent, and writes a real contact and deal into HubSpot or Pipedrive with the transcript attached — or books the meeting. Here is how it works on the website and on WhatsApp, how to write the qualifying questions, and what to measure. A lead capture chatbot is an AI agent that turns a website or WhatsApp conversation into a qualified lead: it answers the visitor's product questions from your own content, asks the qualifying questions you defined, asks for consent, and then writes a contact and a deal into your CRM — HubSpot or Pipedrive — with the full transcript attached, or books the meeting directly. The classic inbound trade-off goes away: a short form captures nothing useful, a long form kills conversion, and either way sales opens the CRM to a name, an email and a mystery. In a conversation the visitor tells you the budget context, the timeline and what they actually need, because the agent is answering their questions at the same time. ### What a lead capture chatbot actually does - Answers first. Pricing, packaging, how it works, whether it fits — from your website and documents, with the source cited, in the visitor's language. - Qualifies in conversation. Use case, timeline, team size, budget band: asked naturally, one at a time, when the conversation has earned them. - Asks consent before storing contact details, in the visitor's language. - Writes to the CRM. A contact matched on email, a deal in the pipeline and stage you chose, the transcript as a note. - Books or hands off. A calendar link, a booking request through your API, or a warm handoff to a rep in the inbox. - Leaves unqualified visitors alone. They still get real answers; no task is created for anyone. ### What lands in the CRM | Object | What the agent writes | | --- | --- | | Contact | Matched on email first — a returning prospect updates the record you already have, never a duplicate | | Deal / Lead | Created in the pipeline and stage you chose, assigned to the owner you picked (Pipedrive: the native Leads inbox; HubSpot: a deal, which exists on every tier) | | Note | The whole conversation, so the first human touch starts informed | ### Why conversation beats forms - Visitors answer questions in context they would never type into a form — because the agent is also answering theirs. - Qualification runs 24/7 and in 12 languages, so the overnight international lead is qualified before your morning coffee. - Unqualified traffic still gets real answers — no follow-up task is created, and no rep time is spent finding that out. - Every channel funnels to one place: the website widget, WhatsApp and Telegram all write to the same CRM with the same rules. ### How to write the qualifying questions The agent asks what you tell it to establish, in plain language in the prompt editor. Three questions is usually the right number; five is a form again. Write them as what your best rep asks in the first two minutes: | Question | Why it qualifies | Example wording | | --- | --- | --- | | What are you trying to do? | Fit — separates a real use case from curiosity | "What would you use this for, mainly?" | | When do you need it? | Timing — a deal this quarter is worth a call today | "Is this something for the next few weeks, or later in the year?" | | How big is the team or the order? | Size — routes to the right rep and the right offer | "Roughly how many people would be using it?" | | Who else is involved? | Authority — optional, only for larger deals | "Would anyone else be part of that decision?" | Then tell it what to do when the bar is met — book, hand off, or write the deal — and what to do when it is not: answer, offer the newsletter or the docs, and stop. The stop is the part most teams forget to write down. ### Consent, and what you are allowed to keep The agent asks before it stores a name, an email or a phone number, and the answer is recorded with the lead. Do not skip this to lift conversion: a lead captured without consent is a lead you cannot legally email, and in the EU the chatbot also has to say it is an AI at the start of the conversation. Both are one-line settings; both are still your responsibility. The transcript stays in your inbox as the record of what was asked and agreed. ### From qualified to booked A qualified lead in a CRM is good; a qualified lead with a meeting on the calendar is better, and the agent can get there in three ways. The simplest is a booking link sent in the thread at the right moment, once the qualifying questions are answered. Where your scheduling or booking system has an API, an HTTP tool lets the agent create the request itself with the details already collected. And where the deal is large enough that a person should choose the slot, the handoff goes to the inbox with the transcript and the rep books it. Pick per pipeline, not per platform — a €50 booking and a €50,000 deal should not follow the same path. ### Lead capture without a CRM Plenty of teams have no CRM yet, or run one the platform does not connect to natively. Leads are still captured, with the same consent and the same qualifying answers: they appear in the workspace inbox and a leads table in the dashboard, fire a lead.captured webhook you can point at n8n, Make, Zapier or your own endpoint, and are readable over the REST API. Connect a CRM later and nothing about the conversation changes. ### On WhatsApp, the same rules The agent that qualifies on your website qualifies on WhatsApp and Telegram with the same questions, the same consent and the same CRM write, because it is one agent on several channels, not a bot per channel. The difference is behavioural: WhatsApp conversations are shorter and more impatient, so keep the qualifying questions to the essential ones and let the rest be collected by the rep. ### What to measure - Conversations that reached a qualifying question — if it is low, the answering layer is weak and visitors leave before the agent earns the ask. - Qualified rate — leads written versus conversations that hit the bar; watch it in both directions, since over-qualifying creates junk deals. - Time to first human reply on written deals — the whole point of the transcript is that this drops. - Meetings booked per qualified lead, per pipeline. ### The honest limits An agent qualifies and routes; it does not run your sales process. It will not chase a rep who ignores the deal, and it should not negotiate custom pricing — configure guardrails so it books the meeting instead. It also cannot qualify against criteria you never wrote down. The goal is that every human minute starts on a qualified, documented conversation, not that humans leave the loop. ### Setting it up 1. **Connect the CRM** — HubSpot or Pipedrive, one OAuth click from the dashboard. Pick pipeline, stage and owner; map your fields. The integration never modifies your schema. No CRM yet? Skip this step — leads land in the inbox and the leads table. 2. **Define qualified** — Tell the agent what to establish — use case, timeline, size — in plain language in the prompt editor, and what to do when the bar is met and when it is not. 3. **Point it at your content** — Import your site and pricing pages so the agent answers accurately while it qualifies; answers cite their source. Sync keeps them current as pages change. 4. **Add the booking path** — A calendar link, an HTTP tool to your scheduling system, or a handoff rule for the deals that deserve a person. 5. **Watch the inbox, tune weekly** — Read the transcripts behind the first week of deals; tighten the questions where the agent over- or under-qualified. It compounds. Plans are priced in EUR with a 7-day trial. Import your site, write your three questions, and read the first twenty transcripts — you will know within a week whether the leads are better than the form's. ## EU AI Act and chatbots: the disclosure rule, and a compliance checklist for customer service https://supercognit.com/blog/eu-ai-act-chatbot-disclosure · 2026-08-24 Article 50 of the EU AI Act has applied since 2 August 2026: if people can talk to your AI, you have to tell them — at the point of interaction, in their language, on every channel. Here is what the rule asks of a customer-service chatbot, what it does not, and a checklist you can finish in an afternoon. Since 2 August 2026 the transparency obligations of the EU AI Act apply to any chatbot that talks to people in the EU. The headline rule fits in a sentence: if a person is interacting with an AI system, they have to be told — at the moment the conversation starts, not in a policy page. Most websites running a chat bubble have not changed a thing, which is why this is worth twenty minutes of your week. This guide covers what Article 50 asks of a customer-service chatbot, the exception that is thinner than it looks, who is responsible when an agency runs the bot, and a compliance checklist you can work through in an afternoon. ### What Article 50 actually asks for The obligation is that people are informed they are interacting with an AI system, at the point of interaction. Not in a policy page, not in a footer — at the moment the conversation starts. A parallel obligation covers marking AI-generated content in a machine-readable way; generative systems already on the market have until 2 December 2026 to comply with that part. ### The obvious exception is thinner than it looks Disclosure is not required where it is obvious from the point of view of a reasonably well-informed, observant and circumspect person that they are dealing with AI. Published guidance cautions against leaning on that. A chat panel in the corner of a shop that answers instantly at three in the morning is obvious to you, because you built it. It is not obvious to a sixty-year-old buying garden furniture. ### It binds more people than you would expect Providers have to design the disclosure in. Organisations deploying somebody else's system cannot assume it has been handled for them — the duty has to sit in their own compliance and governance framework. And the reach is territorial by user, not by company address: serving EU users is enough, wherever you are established. ### When an agency runs the chatbot for you A common arrangement is that an agency builds and operates the chatbot on a client's website. That does not move the obligation: the business whose customers are talking to the bot is the deployer, and it is the deployer's compliance framework the disclosure has to live in. The practical answer is to write it into the engagement — who owns the wording, who checks it survives a redesign, who keeps the transcripts — so that nobody discovers the gap during a complaint. ### What the AI Act does not require from a customer-service chatbot - A customer-service chatbot that answers questions and captures enquiries is not, on its own, a high-risk system; the heavy obligations of the Act attach to specific high-risk uses, not to talking to customers. - It does not require a human to review every answer, or an approval process before each conversation. - It does not require you to stop using AI-generated replies — only to be transparent about them. - It does not replace the GDPR. Consent before storing contact details, and a lawful basis for follow-up, are separate duties that still apply. This is a plain-language summary, not legal advice; your counsel should confirm how the Act applies to your specific use. ### What good compliance actually looks like - Say it in the first message the visitor sees, not in a link they will not click. - Make it survive language switching — a disclosure that only exists in English is not a disclosure for a Portuguese visitor. - Keep it visible on every surface the agent runs on: widget, hosted page, WhatsApp, Telegram. - Keep a record of what was disclosed and when, which mostly means keeping your transcripts. - Mark AI-generated content where you publish it, ahead of the December date. ### A compliance checklist for customer-service chatbots 1. **Put the disclosure in the welcome message** — One sentence, first message, before the visitor types. "You are chatting with an AI assistant; a person can take over at any time" does the job. Avoid names and avatars that imply a human. 2. **Check it in every language the bot speaks** — Open the chat as a Portuguese, Spanish and German visitor. If the agent follows the visitor's language, the disclosure must follow too. 3. **Check it on every channel** — Website widget, hosted chat page, WhatsApp, Telegram. A disclosure that lives only in the widget's header is missing on the channels where people cannot see a header. 4. **Make the handoff to a human explicit** — When a person takes over the thread, say so. The rule is about not misleading people in either direction. 5. **Keep the transcripts** — Being able to show what your agent said, to whom, and when is the boring core of compliance. Retention should match your privacy policy. 6. **Separate consent from disclosure** — Telling someone they are talking to an AI is not consent to store their email. Ask for that separately, before storing anything, and record the answer. 7. **Assign an owner** — Someone has to check the disclosure is still there after the next website redesign, the next agent rewrite, and the next channel you add. Write the name down. ### Where SuperCognit already helps The welcome message and behaviour settings are edited visually, so adding a line is a two-minute change rather than a ticket. The chat interface ships in 12 languages and the agent follows the visitor's language, so the disclosure travels with them across the widget, the hosted page, WhatsApp and Telegram. Consent is asked before contact details are stored, and every conversation lands in the workspace inbox — which is the record you will need if anyone ever asks what your agent said. Of all the AI regulation arriving over the next two years, this is the cheapest to satisfy and the most embarrassing to be caught missing. The deadline has passed; the checklist takes an afternoon. ## WhatsApp's other price change: AI replies are now billed by the token https://supercognit.com/blog/whatsapp-token-pricing · 2026-08-24 From 1 August 2026 Meta bills Business Agent conversations on tokens consumed, and from 1 October it starts charging for service replies again. Two different meters, five weeks apart, and most of the coverage only mentions one. We have written twice about the 1 October change, when free-form replies inside the 24-hour customer service window stop being free. That is the one everybody is planning for. There is an earlier change that got much less attention, and it is the more interesting of the two, because it changes the shape of the bill rather than its size. ### Two meters, five weeks apart From 1 August 2026, Meta Business Agent conversations are billed on tokens consumed rather than per conversation — reported at around two dollars per million tokens, roughly four to five cents per conversation, bundling the AI processing and the message delivery into a single rate. It applies across WhatsApp, Messenger and Instagram. Then from 1 October, service messages are charged per message again, at the per-message rate Meta already applies to utility and authentication templates, set by market. ### Why token billing behaves differently Per-conversation pricing is predictable: you can multiply it by your volume and put the number in a spreadsheet. Token pricing scales with how much the model reads and writes. A long system prompt, a generous slice of retrieved knowledge and a chatty answering style are now line items. Your bill becomes partly a function of your prompt design, which is a genuinely new thing for most teams to have to think about. ### What to do before 1 October - Measure how many service replies you actually send inside the 24-hour window today. Most teams have never counted, and it is usually more than they guess. - Look at your prompt and your retrieval settings as cost inputs, not just quality inputs. - Put a cost cap somewhere in the stack that stops spending rather than quietly absorbing it. - Keep at least one channel you own — a widget or your own domain — where the only meter is the model's. ### The structural point None of this is Meta behaving badly; running a messaging network costs money. But it is a reminder of what renting a channel means. When the channel is somebody else's, the meter is somebody else's too, and it can change twice in one quarter. An agent that also answers on your own domain and inside your own site does not get repriced by a conference announcement. The specifics above were announced at Meta's Conversations event in June 2026 and reported widely since; rates are set per market, so check the figures against your own rate card before you build a forecast on them. ## Most MCP servers still skip OAuth. That is the whole security story. https://supercognit.com/blog/mcp-oauth-gap · 2026-08-24 The specification makes OAuth 2.1 mandatory. Surveys of the public server ecosystem suggest only a small minority implement it. If you are about to connect an MCP server to an assistant that can see your work, that is the first thing to check. The Model Context Protocol has had a very good eighteen months. Thousands of public servers, adoption in every major AI client, and a specification that now moves through an open process. The security picture is less flattering: the spec requires OAuth 2.1, and third-party surveys of public servers keep finding that only a small fraction actually implement it. ### What the specification requires OAuth 2.1, and as of the July 2026 revision, a noticeably harder version of it. Authorization servers should return the iss parameter per RFC 9207 and clients must validate it before redeeming a code, which closes an authorization-server mix-up attack. Client credentials are bound to the issuer that minted them, so they cannot be replayed against a different authorization server. And Client ID Metadata Documents are now the preferred registration path — Dynamic Client Registration is formally deprecated in favour of CIMD, though it keeps working for compatibility. ### Why a buyer should care - A server with no authentication that touches real data is an open endpoint with a friendly description. - Credentials that are not issuer-bound are a confused-deputy problem waiting for someone to notice. - Without per-seat identity there is no way to revoke one person's access — only everyone's. - If a server cannot tell you who called a tool, it cannot tell you who leaked something either. ### What we do SuperCognit's MCP endpoint implements OAuth 2.1 including issuer validation and CIMD registration. Tools list without signing in, deliberately, so a prospective buyer can paste the URL and see exactly what a server offers before any credential changes hands; OAuth kicks in on the first actual tool call. That combination — inspectable without an account, authenticated before it does anything — is what we would want from somebody else's server. If you are evaluating a server, the checklist is short: does it do OAuth at all, does it validate the issuer, can you revoke a single seat, and does it log who called what. A server that fails the first question does not need the other three. ## MCP Shadow IT: your team is already connecting servers nobody vetted https://supercognit.com/blog/mcp-shadow-it · 2026-08-24 Connecting a remote MCP server takes a URL and about ten seconds. That is the feature. It is also why a new category of shadow IT now has a name and a threat model. Every wave of shadow IT has the same shape: a tool that is useful, free to start, and does not need anyone's permission. SaaS took a credit card. Browser extensions took two clicks. MCP servers take a URL pasted into a config file, and the assistant on the other end already has the employee's context loaded. ### Why this spreads faster than the last one There is no procurement step, no install, no admin rights, and frequently no sign-in at all. The friction that used to give security teams a chance to notice simply is not there. And unlike a rogue SaaS account, an MCP server is not a silo off to the side — it is wired into the assistant that is already reading the work. ### What actually goes wrong - The server sees whatever the model passes it, and the model is not careful about what is relevant. - Tool descriptions are instructions the model reads, which makes every third-party server a prompt-injection surface. - There is usually no audit trail on the company side — you learn what happened from the vendor, if at all. - Credentials get handed to a party nobody assessed. - A server can change what it does after you approved it, and nothing tells you. ### Questions worth asking before a server is allowed - Who operates it, and is that a company or a weekend project? - Does it implement OAuth, and can you revoke one person without revoking everyone? - What does it log, and can you get those logs? - How would you find out if its tools changed? ### The fix is usually supply, not prohibition People connect random servers because they need something they cannot get otherwise. Blocking the category buys a few weeks. Publishing one sanctioned server that does the lookups they were reaching for — invoices, stock, ticket status, whatever your team keeps asking each other — removes the reason. That is a weekend of work on a platform where tools and knowledge are workspace-level and the endpoint is behind your own sign-in. ## MCP went stateless. What changed, and why we still speak the old protocol. https://supercognit.com/blog/mcp-stateless-spec · 2026-08-24 The 2026-07-28 revision removed the initialize handshake and the session header. It makes MCP servers scale like any other HTTP service — and it leaves every client that has not caught up behind. The July 2026 revision of the Model Context Protocol is the largest change since the protocol was published, and it is easy to under-read. It turns a bidirectional stateful protocol into a request and response stateless one. If you operate a server, this is the release that changes your architecture. ### What was removed The initialize and initialized exchange is gone, and so is the Mcp-Session-Id header. Each request now travels on its own, carrying its protocol version, client identity and client capabilities in _meta. The practical consequence is stated plainly in the spec: any request can land on any server instance behind a plain round-robin load balancer, without shared storage. ### Why that matters operationally Sessions meant sticky routing or shared session state, which is the thing that makes a service annoying to scale and annoying to deploy. Removing them turns an MCP server into an ordinary stateless HTTP service that you can run several of and restart whenever you like. For anyone who was dreading the operational story of running MCP in production, this is the good news release. ### Where state goes instead Application-level state does not disappear, it moves up a layer: a tool mints an explicit handle and the model passes it back as an argument on the next call. State becomes something you design deliberately rather than something the transport gives you for free, which is generally how it should have been anyway. ### The compatibility problem, and what we did about it The spec commits to a minimum twelve-month deprecation window, and deprecated features keep working through it. Clients in the wild move more slowly than specifications do. So our MCP server speaks both eras: the current stateless specification and the legacy handshake, chosen per request. We expect to keep the legacy path until around August 2027, because the alternative is silently breaking somebody's integration on a Tuesday. If you run a server, the temptation is to ship the new era and delete the old one the same week. The spec gives you twelve months for a reason. Use them. ## MCP is not one vendor's protocol any more https://supercognit.com/blog/mcp-linux-foundation · 2026-08-24 Anthropic donated the Model Context Protocol to the Agentic AI Foundation under the Linux Foundation in December 2025. If lock-in was your objection to building on MCP, it no longer holds. The most common objection we heard about MCP in its first year was not technical. It was that the protocol belonged to one AI company, and building your integration surface on a competitor's standard is a bet you cannot easily unwind. That objection has an answer now. ### What happened In December 2025 the protocol was donated to the newly formed Agentic AI Foundation, under the Linux Foundation. Membership has grown steadily through 2026, reported in the hundreds by the middle of the year. More importantly than the headcount, the specification now moves through a public process with a published roadmap and an explicit deprecation policy. ### Why governance is the part that matters A protocol owned by a single vendor is a roadmap risk: the thing you built against can be repriced, deprioritised or quietly folded into a product tier. Foundation governance plus a minimum twelve-month deprecation window is not a guarantee of anything, but it is the difference between planning and hoping. It is why we were comfortable making MCP a first-class product on our platform rather than an integration. ### What it does not change Standardisation is not safety. The same ecosystem that now has open governance also has a large share of public servers that skip the specification's mandatory OAuth. A standard tells you what good looks like; it does not make anybody implement it. For a buyer, the practical upshot is that publishing an MCP server is no longer a bet on one client winning. You publish once and every compliant client can connect — which is exactly the property you wanted from a standard in the first place. ## Who pays when the agent buys? AP2, x402, ACP and MPP, plainly https://supercognit.com/blog/agent-payments-protocols · 2026-08-24 Four protocols, four layers of the same problem: letting software spend money on your behalf without it becoming a disaster. Here is what each is actually for, and what is still genuinely unsettled. Agentic commerce is the idea that an agent takes a goal rather than a click and runs the whole purchase — discovery, authorization, payment, fulfilment. The interesting engineering is not the buying. It is proving that the human meant it, and deciding who is liable when the machine got it wrong. ### The four names you will keep hearing - AP2, the Agent Payments Protocol — introduced by Google Cloud and Coinbase in September 2025 with a large partner list. It is the authorization and trust layer: cryptographically signed mandates that record what the human actually authorised. Supports card rails, and crypto through an x402 extension. - x402 — from Coinbase, reviving the long-dormant HTTP 402 status code for stablecoin payments over HTTP. Built for machine-to-machine microtransactions and API monetization. It has the most production traction of the four: V2 landed in December 2025, Stripe integrated it in February 2026, and Cloudflare supports it. - ACP — the checkout layer for conventional e-commerce flows. - MPP — machine-to-machine payments, in the same neighbourhood as x402. ### They are layers, not rivals The framing that trips people up is treating this as a format war. A real system might use AP2 to capture and verify the authorization, ACP to complete a retail checkout, and x402 or MPP for the machine-to-machine calls underneath. Asking which one wins is like asking whether OAuth beat HTTP. ### What is genuinely unsettled Security research on x402 is already being published, including systematic analyses of attacks against it — which is healthy and also a signal of maturity. Signed mandates raise a harder question than the cryptography: what does a signed intent mean when the model misread the goal it was given? And nobody has a clean answer for disputes and refunds when the counterparty was a machine acting under delegated authority. ### Where we stand Fiat first. An agent that can raise a payment request inside a conversation and have it reconciled through the platform's existing billing covers the overwhelming majority of what our customers actually need today — a consultant taking a deposit, a shop closing a sale in chat. We are watching the protocols closely and we are not asking anyone to bet their checkout on one of them yet. ## The SuperCognit AI API: build, publish and monitor agents from your own code https://supercognit.com/blog/supercognit-ai-api · 2026-08-24 Everything you do in the dashboard, you can also do over HTTP: create agents and MCP servers, attach tools and knowledge, publish them, read every transcript and receive webhooks. Here is the whole surface — and the first call you will make. SuperCognit is a no-code platform, and most people never need anything else: describe the agent, attach your knowledge, publish. But the moment you are running agents for a hundred customers, wiring conversations into a CRM, or building SuperCognit into your own product, clicking stops scaling. That is what the API is for — the same workspace, driven by your code. ### One key, one header Create an API key in your workspace and send it as an X-Api-Key header. Keys declare scopes — write is required for anything that is not a GET — and a key can be locked to a single product, in which case it cannot touch workspace-level resources such as tools, knowledge bases or webhooks. That is the whole authentication story: no OAuth dance, no SDK to install. ``` curl https://api.supercognit.com/v1/products \ -H "X-Api-Key: $SUPERCOGNIT_API_KEY" ``` ### Create and publish an agent without opening the dashboard A product — a chat agent or an MCP server — is created as a draft, configured, and then published. That is exactly the lifecycle the dashboard follows, which is why agencies use it to provision one agent per client straight from their own onboarding flow. ``` curl -X POST https://api.supercognit.com/v1/products \ -H "X-Api-Key: $SUPERCOGNIT_API_KEY" \ -H "content-type: application/json" \ -d '{ "kind": "chat_agent", "title": "Acme support", "intro": "Hi! Ask me anything about Acme.", "systemPrompt": "You answer Acme customer questions from the attached knowledge." }' ``` The response carries the new product's id and slug. From there you attach tools and knowledge bases, patch the prompt or the price, and publish it — all over the same API. ### What the API covers - Products — create, update, publish and unpublish chat agents and MCP servers. - Tools — the workspace tool library: LLM, HTTP and static tools, their inputs, and which products use them. - Knowledge bases — the knowledge behind your answers, and where each base is attached. - Conversations — sessions and full transcripts, with the model and token counts on every message. - Webhooks — endpoints, their delivery log and their retries. - Wallet — the usage-credit balance and every top-up, charge and refund. - End users — verify a chat session token server-side, so your app knows who it is talking to. ### Events, pushed to you Polling is fine for backfill, but realtime should arrive on its own. Register an endpoint and SuperCognit POSTs JSON as conversations happen: conversation.created when someone starts a conversation, message.created for each message, message.completed when the agent finishes its answer. Deliveries are retried up to five times with exponential backoff and logged with their response status, so nothing disappears quietly. Every request is signed. Compute the HMAC-SHA256 of the RAW body with your endpoint secret and compare it to the X-Supercognit-Signature header — against the exact bytes you received, never a re-serialized copy of the JSON. ``` import { createHmac, timingSafeEqual } from "node:crypto"; const expected = "sha256=" + createHmac("sha256", WEBHOOK_SECRET).update(rawBody).digest("hex"); const valid = timingSafeEqual(Buffer.from(expected), Buffer.from(signatureHeader)); ``` ### What teams build with it 1. **An agent per customer** — Your onboarding creates the product, attaches the right knowledge base, publishes it and hands the customer a working agent — before your welcome email arrives. 2. **Conversations in the CRM** — A webhook per message pushes conversations and captured contacts into HubSpot, Salesforce or your own database, where the rest of your funnel already lives. 3. **Knowledge that follows the product** — When your documentation changes, refresh the source instead of waiting for the next scheduled crawl — answers stay in step with what you actually ship. 4. **Usage you can bill on** — Read the wallet and the transcripts to see exactly what an account consumed, then charge for it inside your own product with your own margin. ### REST for your systems, MCP for AI clients The two developer surfaces answer different questions. The API is how your code drives SuperCognit. MCP is how AI clients — Claude, Cursor, your own agents — use what you publish. Both run on the same workspace, so a tool you build once is reachable from either side, and the platform's own MCP server exposes the same resources inside the assistant you already work in. ### Start with the reference The full reference is public and interactive at api.supercognit.com/docs, generated from an OpenAPI 3.1 document you can feed straight into a client generator. Create a key, run the first call above, and watch your workspace answer in JSON — free for 7 days. ## From consultant to product: packaging your expertise as a paid MCP server https://supercognit.com/blog/consultant-to-product-paid-mcp · 2026-08-21 Packaging your expertise as a paid MCP server means turning the material you already deliver into a knowledge base clients query from inside Claude — with per-seat billing built in, and no code to write. Packaging your expertise as a paid MCP server means taking the material you already deliver to clients — the frameworks, the checklists, the answers to the same forty questions — and publishing it as a knowledge base they can query from inside Claude or any MCP client, with access sold per seat. On SuperCognit the packaging itself takes no code: upload the documents, import your site, publish, set a price. The hard part is deciding which of your expertise is repeatable. That question is worth answering even if you never ship the product. ### The ceiling you already know about Consulting sells hours, and hours do not scale — you knew that when you started. The part that stings is subtler: a meaningful share of every engagement is delivery of things you have delivered before. The bespoke judgment is why clients hire you. The repeatable layer is what fills the weeks. The repeatable layer is the product. ### Engagement versus product The comparison is not consulting or product. It is consulting plus a product that handles the layer beneath it. | | Consulting engagement | Paid MCP server | | --- | --- | --- | | What the client gets | Your calendar, in scheduled blocks | Your documented expertise, on demand, inside the tools they already work in | | Marginal cost of one more client | More of your week | A seat | | When questions get answered | On the next call | Mid-task, with the source cited | | What you maintain | Each engagement separately | One knowledge base, re-synced on a schedule | | How it bills | Invoices you write and chase | Per-seat Stripe subscriptions, invoiced and paid out automatically | ### Who this fits Any consultant whose clients ask questions between calls. A compliance specialist whose clients are 40-person accounting firms, each with staff who need the same rules explained mid-file. A hospitality consultant whose boutique-hotel clients retrain front-desk staff every season. An operations consultant whose SOPs get implemented, forgotten and reinvented. In each case the client's team already works in tools that speak MCP, and your knowledge can meet them there instead of waiting for the next workshop. ### What expertise looks like as a knowledge base Anything you have written twice belongs in it. In practice: - Methodology documents and playbooks — the thing you walk every new client through. - Checklists and decision guides — judgment you have already codified without noticing. - Worked examples with the reasoning attached, not just the outcome. - The explanations you keep rewriting in email, promoted to documents at last. Import your site as well — articles you have published are crawled into the knowledge base in about two minutes — and everything is cited: when a client's team asks a question mid-task, the answer names the document it came from. Citations are the difference between a product and a rumor. ### The mechanics you don't build The unglamorous parts are the reason this was not viable to do alone. Auth: every server implements OAuth 2.1 to the current spec, with issuer validation and CIMD registration, so it connects cleanly from Claude and any other MCP client. Isolation: your product lives in its own workspace and answers only from it. Billing: per-seat Stripe subscriptions, invoices and payouts are built in — SuperCognit is the merchant of record and takes a platform fee, and your side of the transaction is a payout. Brand: custom domains, so clients connect to your name rather than a platform's. The same build also produces a chat agent, which turns out to be the sales asset: a prospect can talk to the product before anyone buys a seat. ### What it doesn't replace The server handles the repeatable layer; it does not do the judgment, and you should not pretend it does. What actually happens is cleaner: subscribers who outgrow the server's answers are exactly the clients worth a proposal, and they arrive already convinced your material is good. The product does not cannibalise the consulting. It qualifies for it. ### Start smaller than feels serious One methodology, a dozen documents, a handful of seats sold to clients who already trust you. Knowledge re-syncs on a schedule, so the product improves every time you improve the material — which you were doing anyway. Plans are in EUR with a 7-day trial; the trial is long enough to find out whether your repeatable layer is as documented as you think it is. ## An MCP server for your agency's clients: one per client, on their brand https://supercognit.com/blog/mcp-server-for-agency-clients · 2026-08-19 The pattern that works for agencies offering MCP servers to clients: one server per client, built from their website and documents, published on their own domain — no code, isolated workspaces, billable per seat. If your agency wants to offer clients an MCP server, the pattern that works is one server per client, built from that client's own website and documents, published on that client's own domain. On SuperCognit this takes no code: import the client's site — it is crawled into a cited knowledge base in about two minutes — upload the documents the site doesn't cover, attach any tools, publish. Each server lives in its own workspace, isolated from every other client you run. One build produces two surfaces. The client gets a chat agent they can put on their website, and an MCP server their team or their customers can add to Claude or any other MCP client. Same knowledge, same configuration, answered in both places — which means the deliverable you demo in the first meeting and the deliverable their team connects to are the same object. ### Why one per client, not one server with client folders The tempting shortcut is a single agency server with a section per client. Resist it. Two clients' pricing rules sharing one knowledge base will eventually meet in one answer, and that is a phone call you do not want to take. - Isolation is structural. A workspace is a hard boundary; a server can only answer from what its own workspace holds. No prompt engineering is doing the separating. - Auth is per server. Every server implements OAuth 2.1 to the current spec — issuer validation, CIMD registration — so who can connect is decided client by client, not agency-wide. - Offboarding is clean. If a client leaves, you hand over or retire one workspace. No other client notices. - The brand separates too. Custom domains are supported, so each server answers at the client's own hostname, not yours. ### What goes into a client's server Three kinds of material, in whatever mix the client's business needs: - The website crawl. Imported and cited, so every answer points back to the page it came from — which matters when the client's own staff are checking the machine's work. - Documents. The things the site never says: rate sheets, policies, the onboarding guide, the FAQ that lives in someone's sent folder. - Tools. HTTP tools you configure become callable MCP tools, so the server can check availability or look up an order instead of only describing how one would. Then one decision per client: private or public. A boutique hotel might publish a public server so a guest's assistant can ask about rooms and rates. A 40-person accounting firm probably wants a private one — an internal brain of SOPs and policies that only staff can connect to. Same build either way; different audience, different boundary. ### The retainer question An agency deliverable that decays is a liability, and most AI deliverables decay fast. This one maintains itself: knowledge re-syncs on a schedule, as often as hourly, so when the client updates their site, the server follows without a ticket. That is what turns the engagement from a project into a productised service you can describe honestly in a proposal. For money, two roads. Fold the server into the retainer you already bill, or sell access per seat directly: Stripe subscriptions, invoices and payouts are built in, with SuperCognit as merchant of record taking a platform fee. Either way you are not building billing infrastructure, and you are not chasing invoices. ### The first client, in an afternoon 1. **Pick the client whose site is the product** — Hotels, clinics, firms with deep service pages — businesses whose website already answers questions are the fastest wins. Four Seasons Fairways, a SuperCognit customer, is exactly that shape: a property whose guests ask the same questions all season. 2. **Import and read** — Crawl the site, then actually read what the knowledge base holds. The gaps you find are the documents to request in your kickoff email — clients respond faster to a specific list than to a request for content. 3. **Publish both surfaces** — The chat agent gives the client something to see in the first meeting; the MCP server is the part their team connects to Claude. Put it on their domain before the handover demo, not after. 4. **Set the sync schedule and step back** — Frequent for pages that change — rates, availability, menus — slower for the rest. Then the maintenance line in your proposal is true without being expensive. Plans are in EUR with a 7-day trial — long enough to build the first client's server before you pitch it, which is the demo that closes it. ## 10 examples of company MCP servers (and what each one answers) https://supercognit.com/blog/company-mcp-server-examples · 2026-08-17 Ten examples of company MCP servers — private internal brains, public knowledge servers and paid per-seat products — with the questions each one answers. A company MCP server is a service that lets AI clients — Claude most visibly, but any MCP client — answer questions and take actions using your company's actual knowledge and tools, with citations. What that looks like in practice varies more than the definition suggests, so here are ten concrete patterns, grouped by who they serve: your own team, the public, or paying subscribers. All ten are built the same way on SuperCognit — import a website or upload documents, attach tools, publish — and each build produces a chat agent and an MCP server from the same knowledge. The companies below are illustrative types, not client names, with one exception flagged where it appears. ### Private: the internal company brain #### 1. The 40-person accounting firm's SOP server Engagement checklists, filing calendars, review procedures, who signs off on what. It answers the questions seniors are tired of answering: what's the onboarding sequence for a new client, which template applies, what's the deadline for this filing type. New hires stop interrupting and start citing the manual. #### 2. The people-ops handbook Leave policy, expense rules, benefits, onboarding steps. Answers "how many days can I carry over" with the actual clause, cited, instead of a colleague's recollection of it. Private by design — handbooks contain things you would never publish, which is the point of a private server. #### 3. The IT runbook VPN setup, device policy, password procedures, which system takes which ticket. Pair the knowledge with an HTTP tool that opens a ticket and the server crosses from answering to acting — agent HTTP tools are callable as MCP tools, so "report this" becomes something the client can do, not just describe. #### 4. The agency's account brain Brand guidelines, tone rules, process documents, per-client conventions. The questions it kills are the expensive ones: the details that live in a senior account manager's head and leave with them. ### Public: so AI answers correctly on your behalf #### 5. The boutique hotel's guest server Rooms, spa hours, cancellation terms, directions, what's open in November. Guests increasingly ask their AI assistant before they email the front desk; a public MCP server means the assistant answers from your pages instead of guessing from the open web. Four Seasons Fairways — the one customer we can name here — is a resort working in exactly this pattern, from the same build that serves its own chat. #### 6. The SaaS documentation server Product docs and plan facts, crawled and re-synced on a schedule — as often as hourly, so a shipped change reaches answers the same hour. Customers connect it in their own MCP client and get cited answers instead of filing tickets. #### 7. The manufacturer's spec server Spec sheets, compatibility tables, installation guides. Distributors and installers ask "does part A fit housing B" all day; a server built from the actual PDFs answers with the page it read. #### 8. The property manager's building server House rules, maintenance procedures, tenant FAQs, per building. Add a maintenance-request tool and a tenant's assistant can both look up the rule and file the issue. ### Paid: knowledge with a price on it #### 9. The consultancy's research server A firm that sells sector analysis publishes its research corpus as an MCP server and charges per seat — Stripe subscriptions, invoices and payouts are built in, with SuperCognit as merchant of record. Clients query the corpus from inside their own AI tools, and a cancelled seat is revoked access. #### 10. The training body's curriculum server Course catalogues, prerequisites, exam policies, certification requirements — sold per seat to member firms whose staff ask compliance questions all year. The content already existed; the server is a new distribution channel for it. ### What the ten have in common Strip the industries away and it's one shape, three scopes. A knowledge base built from a website crawl — about two minutes — or from uploaded documents. A decision: private, public, or per-seat paid. Optionally, tools. The security posture, OAuth 2.1 and per-workspace isolation, comes standard rather than per project. The honest test for whether your company has an MCP server in it: is there a set of questions your organisation answers over and over, from documents that already exist? If yes, one of the ten above is approximately you. ## Migrating from WhatsApp Business: the playbook for the per-reply pricing era https://supercognit.com/blog/migrating-from-whatsapp-business · 2026-08-15 From October 1, 2026, WhatsApp charges businesses for every service message — the replies inside the 24-hour customer window that used to be free. Here's what changes, what it costs you, and a practical playbook to move conversation volume to channels you own. If your business answers customers on WhatsApp, your cost model changes on October 1, 2026. Before that date, replying to a customer inside the 24-hour customer-service window was free — only template messages you initiated carried a fee. From October 1, Meta begins charging for service messages too: every free-form reply your business (or your chatbot) sends inside that window is billed. Customers still message you for free; answering them is what now costs money. ### What exactly changes from 1 October 2026 - Service messages — free-form replies within the 24-hour customer window — stop being free and are charged per message sent. - High-volume support and chatbot flows are hit hardest: every answer, clarification and follow-up in a conversation is a billed message. - Template message pricing (marketing, utility, authentication) continues on its own per-message rates. - AI features on Meta's side are metering too — Meta's Business AI is priced per token on top of message fees. The direction is clear: the more of your customer conversation lives inside WhatsApp, the more your support bill scales with your success. A busy agent that exchanges 20 messages per conversation now has 20 line items where it used to have zero. ### The move: own your primary channel, keep WhatsApp at the edge We're not going to tell you to abandon WhatsApp — it's where many of your customers are, and SuperCognit itself ships a WhatsApp channel. The play is about where the volume lives. An AI concierge on your own website and domain answers unlimited conversations with no per-message channel fee; WhatsApp stays as a doorway, not the whole house. ### The migration playbook 1. **Stand up your own agent first** — Import your website into SuperCognit — the crawler builds a cited knowledge base and drafts your agent in minutes. This becomes your primary, fee-free answer channel. 2. **Put it everywhere your traffic already is** — Embed the widget on your site with one script tag, or give the agent its own branded domain. Every conversation that starts here is a conversation WhatsApp never bills. 3. **Redirect the entry points** — Update the WhatsApp links in your Instagram bio, email signatures, QR codes and site buttons to open your web concierge first. Keep WhatsApp for the customers who truly prefer it. 4. **Add Telegram for messenger-native users** — Telegram's Bot API has no per-message business fees. SuperCognit connects the same agent to Telegram, so messenger fans keep a free-to-serve channel. 5. **Keep WhatsApp connected — but deliberate** — Connect WhatsApp as a SuperCognit channel for the conversations that need it. Same agent, same knowledge — but now it's one channel among several, not a single point of pricing pressure. 6. **Watch the numbers** — Every conversation across web, WhatsApp and Telegram lands in one inbox. Compare volumes, watch your next WhatsApp invoice shrink, and tune from data. ### What this looks like in practice Four Seasons Fairways, a resort in the Algarve, runs its guest concierge on its own website — imported from their site and live in minutes, answering in 12 languages around the clock. Guests get instant answers without the resort paying a messaging platform per reply. That's the model: your knowledge, your domain, your economics. > From 1 October 2026, every WhatsApp reply is a metered event. The businesses that win are the ones whose best answers live on a channel they own. > > — SuperCognit Team ### Start this week A first version of your own agent takes an afternoon, not a quarter: import your site, review the draft, publish. SuperCognit is free for 7 days — enough to have your fee-free channel live before your next WhatsApp invoice arrives. If you'd like help planning the migration, talk to us. ## WhatsApp Business alternatives in 2026: own the conversation instead of renting it https://supercognit.com/blog/whatsapp-business-alternatives · 2026-08-15 With WhatsApp charging for every business reply from October 2026, teams are rethinking where customer conversations should live. Here are the real alternatives — web concierges, Telegram, and an omnichannel AI agent — compared honestly. WhatsApp Business earned its place: your customers are already there, and it works. But its economics keep shifting — from October 1, 2026, businesses pay for every service message they send, including replies your chatbot sends inside the 24-hour window. If a meaningful share of your support or sales runs through WhatsApp, it's the right moment to ask what an alternative actually looks like. ### What you actually need from a channel - Instant answers, 24/7, in your customer's language — the thing customers actually came for. - Your knowledge behind every reply — products, policies, prices — not a generic bot. - Predictable economics — costs that don't scale per reply as you grow. - Ownership — your brand, your domain, your conversation data, exportable and yours. ### The alternatives, compared | | Your own AI concierge (SuperCognit) | WhatsApp Business API | Telegram bot | Email / contact forms | | --- | --- | --- | --- | --- | | Cost per reply | None — plan-based, with cost caps | Charged per message from Oct 1, 2026 | Free Bot API | Free but slow | | Where it lives | Your website & your own domain | Meta's app | Telegram app | Inbox | | Answers instantly, 24/7 | Yes — AI agent with your knowledge | Yes with a paid bot behind it | Yes with a bot | No | | Speaks 12 languages | Yes, automatic | Depends on your bot | Depends on your bot | Manual | | Takes real actions (APIs, bookings) | Yes — HTTP tools, web search, browser | Depends on your bot | Depends on your bot | No | | Owns the customer relationship | You — your domain, your data | Meta's platform terms | Telegram's platform | You | | Discoverable by Google | Yes — it's your website | No | No | No | | Setup time | Minutes — import your website | API onboarding + templates | Bot development | Already there | ### Alternative 1: an AI concierge on your own site The strongest alternative isn't another messenger — it's making your own website the best place to get an answer. A SuperCognit agent learns your business from your website and documents, answers with citations in 12 languages, and can call your APIs to actually do things. It lives on your domain, costs the same whether it answers ten conversations or ten thousand, and every visitor is already there. ### Alternative 2: Telegram For customers who genuinely prefer a messenger, Telegram is the economic outlier: its Bot API carries no per-message business fees. SuperCognit connects the same agent to Telegram natively, so you offer a messenger channel without a metered bill. ### Alternative 3: keep WhatsApp — as one channel of several The honest answer is that WhatsApp still belongs in the mix for many businesses. The mistake is letting it be the only channel. SuperCognit runs the same agent on your website, on Telegram and on WhatsApp simultaneously — one brain, one inbox, and the freedom to steer volume toward the channels with the best economics. > The question isn't 'which messenger replaces WhatsApp?' — it's 'why is my best answer machine living on rented land?' > > — SuperCognit Team ### Try the alternative in an afternoon Import your website into SuperCognit and watch your own concierge come to life — free for 7 days, no code required. If you're planning a move, read our migration playbook or talk to us. ## Introducing SuperCognit: build, publish and monetize AI agents — no code required https://supercognit.com/blog/introducing-supercognit · 2026-08-14 Today we're opening SuperCognit to everyone: one platform to build customer-facing and internal AI agents, publish them on your own domain, and get paid — with knowledge, tools, MCP servers and billing built in. Every company we talk to wants the same thing: AI that actually works for their business — answering their customers, in their languages, with their knowledge — without hiring an engineering team to glue models, vector stores, billing and hosting together. Today we're launching SuperCognit to close that gap. ### What you can build SuperCognit is a workspace where a prompt becomes a product. You describe what your agent does, attach the knowledge it should master, pick the tools it can use — and publish. Everything ships together: - AI agents — customer-facing concierges, support assistants and internal copilots that run a real multi-step tool loop, not a canned Q&A widget. - Knowledge bases — upload PDFs and documents, link Google Docs, or import your whole website; answers come back with citations. - Tools — connect your APIs with the HTTP tool builder, run web search, or automate the browser for the tasks that need real hands. - MCP servers — every workspace can expose its tools over the Model Context Protocol, so Claude, Cursor and any MCP client can talk to your product. - An embeddable widget — one script tag puts your agent on any site, preloaded and instant. - Custom domains — your agent lives on your domain with SSL handled for you, not on a subdomain of ours. - Channels — the same agent answers on the web, on Telegram and on WhatsApp. ### Publish it. Then sell it. The part we're proudest of: monetization is native. Set a monthly price, and SuperCognit handles checkout, seats, invoices and payouts. Agencies and experts use it to package what they know into products their clients subscribe to — without touching Stripe dashboards or writing a line of billing code. ### Fast to start, serious underneath You can go from a website URL to a working, branded agent in minutes — our importer crawls your site, builds the knowledge base and drafts the agent for you. Underneath, the platform meters every model call, caps costs per seat and globally, supports 11 end-user languages, and runs prompt caching so production traffic stays affordable. > We wanted the distance between 'I know something valuable' and 'people pay me for an AI product' to be one afternoon. > > — Ignacio Correia, founder ### Start today SuperCognit is live now with a 7-day free trial — bring a prompt, leave with a product on your own domain. If you're an agency or consultant who builds for clients, read about our partner program, or just say hello at hello@supercognit.com. ## We're looking for partners: build AI products for your clients on SuperCognit https://supercognit.com/blog/supercognit-partner-program · 2026-08-14 Agencies, consultants and AI builders: package your expertise into agents and MCP servers your clients subscribe to. Wholesale pricing, a partner portal, and a managed cloud underneath — you keep the client relationship. Since we opened SuperCognit, the most interesting builders on the platform haven't been companies building for themselves — they've been agencies and consultants building for their clients. Today we're formalizing that with the SuperCognit Partner Program. ### The model: you build, we run You know your clients' industries; we run the infrastructure. As a partner you get a portal to manage client workspaces, wholesale seat pricing you can resell at your own margin, and white-glove support. Your clients get agents on their own domains, in their brand — and you get recurring revenue on top of your services. ### What partners are building - Guest concierges for hotels and resorts that answer in the guest's language, day and night. - Lead-qualification agents for real-estate agencies, embedded on listing pages. - Internal knowledge copilots trained on a firm's playbooks, contracts and SOPs. - Custom MCP servers that expose a client's systems to Claude and other AI tools their team already uses. ### What you get - Wholesale pricing — flat per-seat rates for MCP and agent seats, resell at the price you choose. - A partner portal — create and manage client products, seats and invoices in one place. - Skills — package repeatable capabilities once, deploy them to every client. - Co-marketing — launch case studies with us, like the ones on this site. ### Who we're looking for Digital agencies, AI consultants, vertical SaaS founders and system integrators — anyone who ships solutions to clients and wants an AI product line without building a platform. You don't need to write code; you need to know what your clients need. Apply on the partners page or write to hello@supercognit.com with a sentence about who your clients are. We onboard a small number of partners at a time and work closely with each one. ## Building an MCP server: custom development vs. a no-code platform https://supercognit.com/blog/mcp-server-custom-vs-no-code · 2026-08-14 Building an MCP server: an honest comparison of custom development versus a no-code platform, with a side-by-side table and the three cases where writing your own still wins. The short version: build custom when the server's value is bespoke logic — deep integrations, unusual tool behaviour, requirements no platform anticipates. Use a no-code platform when the server's value is your knowledge — documents, policies, product facts — plus tools a straightforward HTTP call can express. Most company MCP servers are the second kind, which is why "we need a developer for this" is usually the wrong starting assumption, and occasionally exactly the right one. The comparison below tries to be honest about both columns. ### What custom development actually involves An MCP server is not a big program. A minimal one is a pleasant afternoon with an SDK. But minimal and company-grade are different projects, and the distance between them is where the cost lives: - Authorization. OAuth 2.1 as MCP specifies it — issuer validation, client registration via Client ID Metadata Documents, token handling. The part teams underestimate most, and where mistakes are quietly serious. - Hosting and operations. TLS, uptime, logging, a domain, someone on the hook when it's down. - The content pipeline. If your server answers from documents, you're building crawling, parsing, retrieval and citations — a search product bolted to a protocol server. - Freshness. That pipeline has to re-run whenever sources change, forever. - Spec churn. MCP's authorization story has already revised more than once. Someone reads the changelog and ships the migration. None of this is exotic. All of it is maintenance that outlives the initial build. ### What a no-code platform does instead On SuperCognit the sequence is: import your website — crawled into a cited knowledge base in about two minutes — upload any documents that don't live at a URL, attach tools, publish. The output is an MCP server that works in Claude and any MCP client, with OAuth 2.1 done to the current spec and every server isolated to its own workspace. The same build also produces a chat agent, which is a real economy: one knowledge base serving two surfaces. Knowledge re-syncs on a schedule, as often as hourly, so the pipeline-forever problem is the platform's rather than yours. If you want to charge for access, per-seat Stripe subscriptions, invoices and payouts are built in, with SuperCognit as merchant of record taking a platform fee. Custom domains are supported. Plans are in EUR with a 7-day trial, which prices "let's find out" at roughly zero. ### Side by side | | Custom development | No-code platform | | --- | --- | --- | | Time to a working server | Days to weeks, depending on scope | Minutes to import, then publish | | OAuth 2.1 (iss validation, CIMD) | You implement and maintain it | Included, kept current with the spec | | Hosting | Your infrastructure, your pager | Hosted; custom domain supported | | Knowledge and citations | You build the retrieval pipeline | Website crawl plus uploads, cited answers | | Staying current | You re-run ingestion yourself | Scheduled re-sync, down to hourly | | Selling access | Stripe integration is on you | Per-seat billing, invoices and payouts built in | | Bespoke tool logic | Unlimited — it's your code | Whatever an HTTP endpoint can express | | Ongoing cost | Engineering time, indefinitely | A subscription | ### When custom wins Three honest cases. Your tools need behaviour a request-and-response HTTP call can't express — long-lived sessions, streamed intermediate state, exotic auth against internal systems. Your compliance regime requires the server on your own infrastructure end to end. Or the server is your product in a deep sense and you want to own every layer of it. If you have the engineering team and one of those holds, build it — the protocol is well documented and the SDKs are good. ### The middle ground most teams miss The choice is less binary than the table implies. On SuperCognit, an agent's HTTP tools become callable MCP tools — so custom logic can live where custom logic belongs, behind an API your team writes, while the protocol surface, authorization, knowledge and billing stay no-code. You write one endpoint, not one server. In practice that covers most of what "we need custom" turns out to mean. The question to ask isn't whether you can build an MCP server — you can. It's whether the differentiated part of yours is the protocol plumbing or the knowledge and tools behind it. For most companies it's the latter, and plumbing is a strange place to spend your engineers. ## How to secure a company MCP: OAuth, workspace isolation and who sees what https://supercognit.com/blog/secure-company-mcp-oauth · 2026-08-12 How to secure a company MCP server in three layers: OAuth 2.1 with iss validation and CIMD registration, hard per-workspace isolation, and explicit private, public or per-seat scoping. Securing a company MCP server comes down to three questions, answered in order. Who can connect at all? That's OAuth. What can a connected client ever reach? That's workspace isolation. And of what's reachable, who sees which part? That's scoping — private versus public, and seats. Get all three right and "we exposed our company knowledge to AI clients" stops being a scary sentence. Miss one, and the other two don't save you. Here is what each layer means in practice, and what to check on any platform that hosts an MCP server for you. ### Layer 1: OAuth 2.1, to the current spec MCP settled on OAuth 2.1 for authorization, and the word current is doing real work in that sentence. The spec has moved, and recent revisions closed real gaps. Two items are worth knowing by name even if you never implement them yourself. Issuer validation (iss) means the client verifies that an authorization response actually came from the authorization server it started with. Without it, a malicious server can play man-in-the-middle games — so-called mix-up attacks — and walk away with a code meant for someone else. Boring, essential, and easy to skip if you hand-roll. Client ID Metadata Documents (CIMD) are how clients identify themselves to your server. Instead of every MCP client pre-registering with every MCP server — which does not scale past a demo — the client presents a URL describing itself, and the server fetches and verifies that metadata. It's the piece that lets your server work with Claude and with MCP clients you haven't heard of yet, without an open-registration free-for-all. SuperCognit implements both, per the current spec, on every published server. If you're evaluating any hosted MCP product, ask about these two specifically. "We support OAuth" is a 2023 answer to a 2026 question. ### Layer 2: workspace isolation Authorization decides who gets in the door. Isolation decides what's behind the door. Every SuperCognit MCP server is scoped to its workspace: its knowledge bases, its tools, its subscribers. A token for one workspace's server is worth nothing at another's, and there is no query, however creative, that retrieves a neighbour's documents — the boundary is structural, not a filter applied at answer time. This matters more than it sounds. Retrieval systems answer from whatever corpus they can search. If the corpus boundary is soft — a tenant ID in a query filter, say — then the security of every answer rests on nobody ever writing one bad query. A hard per-workspace boundary means the interesting failure isn't possible, rather than merely unlikely. ### Layer 3: who sees what With the perimeter and the walls in place, the remaining decisions are yours — and they are policy, not engineering: - Private — the server is an internal company brain: SOPs, policies, product facts, reachable only by your own team. The right default for anything you wouldn't put on your website. - Public — anyone can connect. Right for documentation, product facts, and anything you actively want AI clients answering correctly on your behalf. - Paid — public, but behind a per-seat subscription. Stripe subscriptions, invoices and payouts are built in, with SuperCognit as merchant of record. A seat is an access-control primitive with a price on it: revoke the seat and the access goes with it. The useful discipline is deciding the tier first and the content second. A server that mixes public product facts with internal escalation contacts in one knowledge base has already made its worst mistake, whatever its OAuth looks like. ### What this buys you day to day An employee connects the company server in Claude, authorizes once, and asks how the refund exception process works. A customer on the public server asks what each plan includes — and can't reach the SOPs, because the SOPs live in a different server in a different scope, not behind a prompt that says please don't mention them. Access control by architecture beats access control by instruction. Models follow instructions well, but "well" is not a security boundary. None of this required your team to run an authorization server, rotate keys, or read the spec's changelog. That's the honest pitch: the security layers of a company MCP are exactly the part you want boring, standard, and someone else's full-time job. ## How to keep a company MCP current: syncing knowledge without re-training anything https://supercognit.com/blog/keep-company-mcp-current · 2026-08-10 How to keep a company MCP current without re-training anything: point the knowledge base at your live website and let scheduled re-syncs — as often as hourly — do the maintenance. You don't re-train anything. That's the answer, and it's worth being precise about why. A company MCP server built on a knowledge base doesn't bake facts into model weights. The model reads your documents at answer time and cites what it read. So keeping the server current means keeping the documents current — nothing more. On SuperCognit, the knowledge base re-syncs from its sources on a schedule, as often as hourly, so current is something the system maintains rather than something a person remembers to do. If that sounds too simple, it's because most of the confusion comes from one bad mental model. ### Fine-tuning is the wrong frame People arrive at "our company AI is out of date" carrying a habit from an older workflow: model gets trained, model gets stale, model gets re-trained. That frame makes freshness expensive, so teams ration it. Quarterly updates. A change-request queue for the knowledge base. Someone whose job it is to maintain the AI. A knowledge-base-backed MCP server works differently. When Claude — or any MCP client — asks your server a question, the server retrieves the relevant passages from your documents and the model answers from those passages, with citations. Change the document and the next answer changes with it. There is no training step to schedule, no weights to touch, no version of the model that "knows" last quarter's prices. The freshness problem reduces to a syncing problem. And syncing problems are solvable with a schedule. ### What actually goes stale Worth being concrete about, because it determines your cadence: - Prices and plans — the classic. The website changed; the PDF someone uploaded in March did not. - Policies — cancellation terms, return windows, service commitments. Legal edits these, and nobody tells the AI team, because there shouldn't need to be one. - People and structure — who owns what, who to escalate to. - Product facts — specs, compatibility, what shipped last week. - Operational details — opening hours, seasonal closures, contact routes. Notice the pattern: almost all of it already lives somewhere canonical, usually the website. The staleness is a copy problem. The knowledge base was a snapshot, and snapshots age. ### Point at the source of truth, then sync to it 1. **Import the live site, not an export** — SuperCognit crawls your website into a cited knowledge base in about two minutes. The crawl is the ingestion path, which matters because it can run again — an exported copy can't. 2. **Upload only what has no URL** — Internal SOPs, spec sheets, documents that never touch the web. Uploads are the one category that genuinely needs a human to replace them when they change, so keep this pile small and deliberate. 3. **Set the re-sync schedule** — Crawled sources refresh on a schedule you choose, as often as hourly. Pages that change often earn a tight cadence; the About page does not. 4. **Let your existing edit flow do the work** — From here, updating the MCP server is just updating the website, which your team already does. Marketing edits a page; within the sync window, every connected MCP client answers from the new version. ### Choosing a cadence without overthinking it Hourly sounds like the obvious pick for everything, and for a small site it's fine. But cadence is really a statement about how much staleness you can tolerate per source. Pricing: minutes. Documentation: hours is usually honest. Policy pages: daily catches everything a human legal review actually produces. The goal isn't maximal freshness everywhere; it's that no source's staleness window is longer than the damage a stale answer would do. ### The failure mode to design against The servers that rot are the ones fed by hand. A folder of uploaded PDFs is a knowledge base with no pulse — accurate on upload day, decaying from there, and the decay is invisible until someone gets a confidently cited answer from a superseded document. Citations help here, incidentally: when an answer cites its source, a stale answer at least points at the stale document, so you fix the cause instead of arguing with the symptom. The discipline, then: anything that changes on the web gets synced from the web. Anything uploaded gets an owner and a review habit. Nothing gets pasted in as free text that has a canonical home elsewhere. ### One build, every surface A detail that makes the economics work: the same build produces both a chat agent and an MCP server. One knowledge base, one sync schedule — the answer a visitor gets on your site matches the answer a colleague gets in Claude. You maintain freshness once and every surface inherits it. Which is the whole point: how current your company MCP is should be a property of the plumbing, not a task on somebody's list. ## How to put your MCP server on your own domain (and why it matters for your brand) https://supercognit.com/blog/mcp-server-own-domain · 2026-08-06 You can put your MCP server on your own domain — a subdomain like mcp.yourcompany.com instead of a platform URL. Here is how custom domains work on SuperCognit, and why the URL you hand out is a brand decision. Putting your MCP server on your own domain takes one DNS record: pick a subdomain — mcp.yourcompany.com is the convention forming — point it where the SuperCognit dashboard tells you, and your server is served from your address, certificate handled. Nothing about the protocol changes. What changes is who the server appears to belong to, and for an MCP server that is worth more than it sounds. ### The URL is most of the product surface A traditional product has a homepage, screenshots, a listing somewhere. An MCP server has a URL. That URL is what gets pasted into a client configuration, what appears in your documentation, what one colleague forwards to another with the words 'add this'. There is no packaging around it. When the address is a platform subdomain, the platform is what people see; when it is your domain, the server reads as your infrastructure — because it is. ### The OAuth moment The first time someone connects, the server asks them to authenticate — SuperCognit servers implement OAuth 2.1 per the current MCP specification, issuer validation and CIMD registration included. An authentication prompt is a moment of judgement: people decide in about two seconds whether signing in feels right. A sign-in associated with your own domain matches the thing they believe they are connecting to. A sign-in from an address they have never heard of gets screenshotted and posted in the company chat with a question mark. ### If you charge for it, this stops being cosmetic SuperCognit lets you sell access to an MCP server per seat — Stripe subscriptions, invoices and payouts are built in, with SuperCognit as the merchant of record taking a platform fee. The moment money is involved, the address is part of the pitch. A 40-person accounting firm selling a tax-rules server to its clients is asking them to put an address in their tools and a subscription on their card; an address the firm visibly owns makes both requests easier to say yes to. ### Addresses are load-bearing Once your URL sits in fifty client configurations, it is infrastructure — every configuration that references it breaks if it moves. That argues for getting the address right before you distribute it widely, and right means one you control. The honest version of this argument cuts both ways: a domain you own is a domain you could point elsewhere. That the strongest case for a custom domain doubles as an exit option is, we think, how vendor relationships ought to work. ### What it does not change - Protocol behaviour: tools, knowledge and authentication work identically on either address, in Claude and any MCP client. - Security: OAuth 2.1, issuer validation and per-workspace isolation are the same either way — a custom domain is identity, not armour. - The build: same knowledge base, same tools, same chat agent published alongside. ### Setting it up 1. **Choose a boring hostname** — mcp.yourcompany.com, or something equally dull. It will be typed into configuration files for years; cleverness ages badly there. 2. **Add the DNS record** — The dashboard shows exactly what to create at your DNS provider. Certificates are provisioned and renewed for you; there is nothing to babysit. 3. **Publish, then treat it as permanent** — The server answers on the new address. Put it in the docs, the onboarding guide, the sales one-pager — and never move it. Handing out the platform URL first and migrating later means updating every client config you no longer control. A custom domain does not make a bad server good. It makes a good server look like what it is: a piece of your company's infrastructure that happens to speak MCP, at an address your customers will still trust — and still reach — years from now. ## How to give your MCP server real tools: connecting your APIs, CRM and calendar https://supercognit.com/blog/mcp-server-real-tools-apis-crm · 2026-08-03 To give your MCP server real tools, you describe HTTP endpoints you already have — URL, parameters, auth — and SuperCognit publishes them as callable MCP tools in Claude or any MCP client. No SDK, no server code. You give an MCP server real tools by describing HTTP endpoints you already have — the URL, the parameters, where the credentials go — and attaching them to the server. On SuperCognit, any HTTP tool you build for an agent becomes a callable MCP tool the moment you publish: no SDK, no protocol code, no separate deployment. If your CRM and your calendar have APIs, and they do, they can be tools. ### Knowledge answers; tools act A knowledge base makes an assistant good at questions: what is the refund policy, what does the premium tier include. Tools make it useful for the part after the question — look up this customer's account, check availability on Thursday, log the request. Most real workflows need both halves, and most MCP servers ship only the first, because the second historically required writing and hosting a server. The unglamorous insight of HTTP tools is that the second half mostly exists already. Your CRM has an API. Your calendar has an API. The internal service someone built in 2019 has an API, undocumented but functional. A tool is a described wrapper around one endpoint of one of those. ### The description is the interface The model never sees your code. It sees the tool's name, its description and its parameters, and it decides when to call the tool, and with what, from those alone. 'Look up a customer by email address; returns their plan, renewal date and open tickets' will be called correctly. 'CRM integration' will be called wrongly, or constantly, or both. Writing tool descriptions is the closest thing this workflow has to programming, and it is done in prose. 1. **Pick one endpoint per tool** — A tool should do one job. Look up a contact by email. List open slots for a given week. Create a draft invoice. Narrow tools get called correctly; a do-everything tool forces the model to guess which of its moods you meant. 2. **Describe it for a reader who cannot see the code** — Say what the tool does, what each parameter means, and what comes back. Name the formats and the units. If a parameter is an email address, say so — the description is the only documentation the model gets. 3. **Attach, publish, done** — Attach the tool to the build and publish. It is now exposed as an MCP tool with a typed schema any client understands — and the chat agent from the same build can call it too. Credentials stay server-side in the tool configuration; they never travel to the client. 4. **Test where your team lives** — Connect the server in Claude or any MCP client, ask something that requires the tool, and watch the call. A vague description reveals itself within three questions. Fix the prose, not the endpoint. ### Design rules that save you later - Least privilege: if the tool only reads, put a read-only API key behind it. A tool cannot exceed the credential it carries. - Return what is needed, not the whole record — a customer lookup does not need to ship every field your CRM stores. - Keep writes small and deliberate: a tool that creates a calendar hold is safer than one that can also delete events. - Prefer several narrow tools over one clever one; the model composes them better than it decodes flags. ### What not to expose A useful test: would you give a new hire this capability on day one, unsupervised? Access to the server runs through OAuth 2.1 with per-seat identity, and each workspace is isolated, so the blast radius is contained — but the cheapest safety mechanism is not attaching the dangerous endpoint in the first place. Deletion, bulk export and anything irreversible can wait until the read-only tools have earned their keep. Start with two read-only tools: a CRM lookup and calendar availability. That pair turns a knowledge server into something that answers 'can we meet Thursday, and is this customer on the premium plan' in one exchange — and it ships in an afternoon. ## One MCP, every tool: making your company knowledge available in Claude, Cursor and beyond https://supercognit.com/blog/one-mcp-every-tool · 2026-07-30 MCP is an open standard, so one MCP server makes your company knowledge available in Claude, Cursor and every other MCP client your team uses. Build it once, without code, and stop maintaining a copy per tool. You do not need an integration per AI tool. The Model Context Protocol is an open standard, so one MCP server makes your company knowledge available in Claude, in Cursor, and in any other client that speaks MCP — one build, one URL, every surface your team works in. On SuperCognit that build takes no code: import your website, upload documents, publish. ### The matrix nobody builds Before a standard, connecting knowledge to tools was a multiplication problem: every source times every tool, each pair its own integration project. That matrix is why company knowledge never made it into anyone's assistant — not because it was hard in principle, but because nobody staffs M times N integration projects for material that lives in a wiki. A protocol turns the multiplication into addition: the server is built once, and every client that implements the standard connects to the same endpoint. The economics of the whole problem change shape, which is usually what a good standard does. ### 'Every tool' is doing real work in that sentence Teams do not standardise on one AI client, and they should not have to. Support lives in Claude. Engineers live in Cursor. Someone will adopt a client next quarter that does not exist today. A per-tool integration strategy has to predict those choices; an MCP server does not care about them. Whatever your people use next, if it speaks the protocol, your knowledge is already there. | Approach | Effort | Freshness | Access control | | --- | --- | --- | --- | | Paste documents into each chat | Constant — per person, per conversation | Stale the moment it is pasted | None; the document is now wherever it was pasted | | Custom integration per tool | An engineering project each, maintained forever | As fresh as each pipeline | Reimplemented for every tool | | One MCP server | Built once, no code | Re-syncs on a schedule, down to hourly | OAuth 2.1, per seat, revoked in one place | ### The colleagues who will never install anything Not everyone runs an MCP client, and a knowledge strategy that only serves the technical half of the company is half a strategy. On SuperCognit the same build that publishes the MCP server also produces a chat agent, so the operations manager who will never edit a client configuration asks the same questions against the same knowledge in a browser. One source of truth, two doors in. ### One place to be wrong The quiet benefit is governance. When pricing changes, you change it once — the imported source re-syncs and every connected client answers from the new version after the next scheduled pull. Compare that with the five copies of the old pricing sheet living in five people's chat histories, each pasted in good faith weeks ago. Answers also cite their sources, so when something looks off, the trail leads to a page someone can fix, not to a paste nobody can trace. ### Private or public, same mechanics The server can be private — an internal brain, gated per seat — or public, if the knowledge is meant for the world: a boutique hotel putting its rates, policies and local know-how within reach of any guest's assistant works exactly the same way. You can even sell access per seat, with subscriptions and invoicing built in. The distribution decision is a setting, not a rebuild. ### Getting there Import your website — it is crawled into a cited knowledge base in about two minutes — and upload the documents that never made it online. Publish. Hand the URL to whoever wants it in their client; OAuth 2.1 handles who gets in, per seat, and your workspace is isolated from every other tenant. The strategic point is small but durable: standards remove the obligation to predict winners. You do not have to know which AI client your company will run in two years. You have to know your own knowledge — and put it somewhere all of them can reach. ## The company brain: building an internal MCP for onboarding, SOPs and policies https://supercognit.com/blog/internal-mcp-onboarding-sops · 2026-07-27 An internal MCP server turns your SOPs, policies and onboarding material into a private company brain every AI assistant on your team can query — cited answers, per-seat access, no code. Here is how to build one and keep it current. An internal MCP server is a private, authenticated endpoint that gives your team's AI assistants access to the company's own knowledge — the SOPs, the policies, the product facts, the onboarding material. On SuperCognit you build one without writing code: import the pages, upload the documents, publish, and everyone who signs in gets cited answers from your material inside whatever AI tool they already use. ### The wiki was supposed to fix this Most companies already wrote everything down. The handbook exists. The SOPs exist, in a folder, versioned optimistically. The problem was never writing — it was retrieval at the moment of need. A new hire with a question does not search the wiki; they ask the person next to them, who answers from memory, and the folklore version of the process gets one more generation of transmission. Meanwhile the team started using AI assistants every day, and those assistants know nothing about your company. Ask one about your expense policy and it will answer anyway — fluently, confidently, and from the general internet. That gap, between the assistant your people actually consult and the knowledge you actually wrote, is what an internal MCP closes. ### What goes in - Onboarding: who does what, how the first weeks run, where things live, which acronyms mean what. - SOPs: the operational how-to — how a refund is processed, how an incident is escalated, how the month-end close actually works. - Policies: expenses, leave, security, remote work — the documents people need exactly once a quarter and can never find. - Product facts: specs, pricing rules, the answers support gives — so internal answers match external ones. ### Building it 1. **Import what already exists** — Point SuperCognit at your site or handbook pages and they are crawled into a knowledge base in about two minutes, every answer carrying a citation to its source page. Upload the rest — PDFs, policy documents, spec sheets — alongside. 2. **Keep it private** — Set the server to private. Access runs on OAuth 2.1 per the current MCP specification, and the workspace is isolated: only seats you have granted can connect, and each person authenticates as themselves. 3. **Publish once, get two products** — The same build produces a chat agent and an MCP server. People who live in Claude or another MCP client connect the server; people who just want to ask a question use the chat. One knowledge base, two doors. 4. **Let it stay current** — Imported sources re-sync on a schedule — as often as hourly. When the policy page changes, the brain follows without anyone re-uploading anything. ### Citations are the trust mechanism An internal answer without a source is folklore with better formatting. Every answer from the knowledge base cites the page or document it came from, so the person asking can check, and the person who owns the document discovers it is actually being read. That changes writing incentives too: when the SOP is what the assistant answers from, keeping it accurate stops being a documentation chore and starts being operational. ### Offboarding is a one-seat operation Because access is per seat, revoking one person's access revokes one person — not a shared password everyone has to rotate. That is a small sentence with large compliance implications. The company brain is one of the more sensitive things you will ever put behind an endpoint, and 'we removed their access on their last day' should be a statement you can make truthfully, on the day. ### What it will not do It will not fix an SOP that is wrong, and it retrieves outdated policy exactly as confidently as current policy — the re-sync schedule helps, but only if someone updates the source. The process that exists only in a senior employee's head cannot be imported either; writing it down is still your job. The brain is a retrieval layer, not a documentation programme. A reasonable way to start: collect the twenty questions new hires actually asked in their first month, load the documents that answer them, and see whether the answers hold up. Loading takes an afternoon, and the seven-day trial is long enough to find out. ## How to build a company MCP your whole team can use inside Claude https://supercognit.com/blog/company-mcp-inside-claude · 2026-07-23 How to build a company MCP server your whole team can use inside Claude: one private, cited knowledge base with OAuth 2.1 sign-in, connected once per teammate and kept current from then on. To build a company MCP your whole team can use inside Claude: import your website and upload your internal documents into one knowledge base, publish it as a private MCP server, and have each teammate connect it in Claude once. Sign-in runs on OAuth 2.1, the server is isolated to your workspace, and from that point everyone's Claude answers company questions from the same cited source — not from memory, guesswork, or whoever sits closest to the person who knows. The pattern this replaces is worth naming. Your team already uses Claude, but Claude does not know your company: not your SOPs, not your pricing rules, not the way you actually run projects. So people paste documents into individual chats, get answers of varying vintage, and the real knowledge keeps living in two colleagues' heads. > If the answer lives in one person's head, it is not an asset. It is a queue. ### What belongs in the company brain - SOPs and runbooks — how things are actually done here - Policies: refunds, cancellations, security, leave - Product facts — specs, limits, versions, pricing - Onboarding material, so new hires can ask instead of interrupting - The internal FAQ nobody maintains but everybody needs ### How teams feed Claude company knowledge today | Approach | Freshness | Citations | Access control | | --- | --- | --- | --- | | Pasting files into each chat | Whatever got pasted that day | None | Whoever has the file | | Links to a shared drive | Current, if anyone opens them | None | Folder permissions | | A private MCP server | One shared source; site content re-syncs on a schedule | Every answer cites its source | OAuth sign-in, workspace-isolated | The first two are not wrong so much as unaccountable. Nobody knows which version anyone is working from, and when an answer is bad there is nothing to trace. ### The rollout 1. **Build it once** — Import the website — crawled into a cited knowledge base in about two minutes — and upload the internal documents the site would never contain. No code in any of this. 2. **Keep it private** — Private is the setting that makes it a company brain: your team, authenticated, and nobody else. Public and even paid-access servers exist on the same platform, but that is a different project. 3. **Publish and share the endpoint** — Publishing produces the MCP server. Send the connection details to the team the way you would share any internal tool. 4. **Everyone connects in Claude** — Each person adds the server and signs in through OAuth once. After that it is simply there — Claude can search the company brain mid-conversation and cite what it found. ### The security paragraph, since someone will ask Published servers implement OAuth 2.1 as the current MCP spec defines it, including issuer validation and CIMD client registration, and every server is isolated per workspace. In practice: your people sign in, the auth is the modern spec rather than a homemade approximation, and your workspace's knowledge is reachable only through your workspace's server. This is the part that quietly kills most build-it-ourselves internal tools, and here it is already done. ### Not only Claude MCP is a protocol, not a Claude feature. The same private server works in any MCP client, so a teammate who lives in a different tool connects to the same brain and gets the same cited answers. You standardize the knowledge without standardizing everyone's tooling. The same build also produces a chat agent, if part of the team prefers a plain chat window over an MCP connection. ### Week one Seed it deliberately: the ten questions your team actually asks, answered by the documents you upload. Then let people try to break it. Wrong answers are productive here — every answer carries a citation, so a bad one points at the precise document that needs fixing, and fixing the document fixes it for everyone at once. Website-sourced facts stay current on their own, re-syncing as often as every fifteen minutes. There is a seven-day trial, with plans priced in EUR. A realistic first milestone: by the end of week one, the newest person on the team gets a better answer about your own processes from Claude than from the colleague who has been there five years. Cited, too. ## How to turn your company documents into an MCP server in 10 minutes https://supercognit.com/blog/company-documents-to-mcp-server · 2026-07-20 Turn your company documents into an MCP server in about ten minutes: upload files into a cited knowledge base, choose private or public, publish, and connect it in Claude. No code involved. Here is how the ten minutes actually splits: two to upload your documents and let them index into a knowledge base, two to import your website alongside them if you want it, a minute to decide who gets access, a minute to publish, and the rest to connect the result in Claude and ask it the questions your team answers by hand every week. No code at any point. What you end up with is an MCP server — a live endpoint any MCP client can query — built from files that were sitting in a shared folder this morning. Documents are the densest company knowledge you have and the least reachable. The procedures, the rate cards, the spec sheets — they answer real questions precisely, and nobody can find them. Turning them into an MCP server changes how they are consumed without changing a word of them. ### Gather the right documents first The build is fast enough that preparation is the actual work. Pull together the documents that answer questions people actually ask: - Standard operating procedures and runbooks - Policies — cancellation, refunds, leave, security, anything with rules in it - Price lists and rate cards - Product specs and data sheets - Onboarding guides and internal FAQs Skip the aspirational material. A vision deck makes a poor knowledge base. The best candidates are the documents somebody gets asked about weekly. ### The ten minutes 1. **Upload the documents** — They are indexed into a knowledge base where every answer traces back to the document it came from. Citations are not a nice-to-have; they are how you audit the thing later. 2. **Add your website, optionally** — The same knowledge base can hold your crawled site next to your uploads — the crawl takes about two minutes and keeps its own citations. Documents cover what the site does not say; the site covers what the documents assume. 3. **Choose private or public** — Private means an internal brain your team signs into and nobody else can reach. Public means anyone with an MCP client can connect — and you can charge per seat if the content warrants it. 4. **Publish** — You get an MCP endpoint with OAuth 2.1 authentication already implemented — sign-in, issuer validation, workspace isolation, none of it yours to build or maintain. 5. **Connect it in Claude** — Add the server, sign in, and ask it the question you got asked yesterday. Check that the citation points at the right document. ### Why this beats pasting files into chats Pasting a PDF into a conversation works once, for one person, for one chat. A published server is one source everyone queries: the same documents, the same versions, the same cited answers, with access controlled by sign-in rather than by who happens to have the file. And the identical build gives you a chat agent too, if you want a web-facing surface reading from the same knowledge. ### The illustrative version Take a 40-person accounting firm. Engagement procedures, filing calendars, internal review checklists — currently living in a document tree three partners can navigate. Uploaded and published as a private MCP server, the same material answers a junior's question inside Claude in seconds, with a citation to the exact procedure. The office manager stops being the search engine. Nothing was rewritten; it was republished. ### Keeping it honest Documents change. When they do, replace the file and the knowledge base moves with you. Website sources go further: they re-sync on a schedule, as often as every fifteen minutes, so the crawled half of your knowledge tracks the site automatically. And because every answer is cited, stale content is findable instead of invisible — a wrong answer points at the exact document that needs updating. Plans are in EUR with a seven-day trial. The ten minutes is real, but budget an extra five: the first thing everyone does after connecting is try to catch it out with the trickiest internal question they know, and it is worth watching where the citations land. ## Withnodes is now SuperCognit: one platform to build, publish and monetize AI agents https://supercognit.com/blog/withnodes-is-now-supercognit · 2026-07-17 Withnodes — the platform at withnodes.com where builders created AI agents and MCP servers — is now SuperCognit. Same team, same technology, a bigger mission: everything you built keeps working, and everything you can build just got larger. Here's what changed, what didn't, and what you can create today. If you arrived here looking for Withnodes or withnodes.com — you're in exactly the right place. Withnodes, the platform where builders turned prompts into working AI agents and MCP servers, is now SuperCognit. This isn't a shutdown or a pivot away from what you knew: it's the same team and the same technology, under one name, with a much bigger platform around it. ### Why the change Withnodes started with a simple idea: building an AI agent should be as easy as describing it. As the platform grew — knowledge bases, tools, custom domains, billing, messaging channels, an MCP server per workspace — the name no longer covered what it had become. SuperCognit is the same product grown up: a platform where companies and creators build, publish and monetize AI agents and MCP servers, no code required. ### What changed, and what didn't | | Before (Withnodes) | Now (SuperCognit) | | --- | --- | --- | | Name & domain | Withnodes, withnodes.com | SuperCognit, supercognit.com | | Your account & agents | Created on Withnodes | Keep working — same login, same data | | Existing links & embeds | withnodes.com URLs | Redirect to supercognit.com — nothing breaks | | Agents & chat | Chat agents from your prompt & knowledge | Same, plus tools, channels and memory | | MCP servers | Early support | A standards-compliant MCP server per workspace | | Monetization | Early plans | Native: pricing, checkout, seats and payouts | | Channels | Web | Web widget, custom domains, WhatsApp, Telegram | ### The platform you get today - AI agents that run a real multi-step tool loop — HTTP tools to your APIs, web search, browser automation — not canned Q&A. - Knowledge bases with citations: upload documents, link Google Docs, or import an entire website in minutes. - An MCP server per workspace, so Claude, Cursor and any MCP client can use your product's tools. - An embeddable widget, your own custom domain with SSL, and native WhatsApp and Telegram channels. - Monetization built in: set a monthly price and SuperCognit runs checkout, seats, invoices and payouts. - 11 end-user languages out of the box, with cost caps that keep bills predictable. ### For Withnodes builders: your next steps 1. **Log in at supercognit.com** — Your account, workspaces and agents are all here — the same credentials work. Old links redirect automatically. 2. **Revisit what your agent can do** — The platform has grown since you built: attach HTTP tools, connect WhatsApp or Telegram, turn on web search, give your agent memory. 3. **Put a price on it** — If your agent is valuable, publish it with a monthly price — checkout, seats and payouts are handled for you. 4. **Claim your own domain** — Move your agent from a shared link to your own branded domain with SSL managed by the platform. > Names change; the promise doesn't. Bring a prompt and your knowledge, leave with a product your users can pay for. > > — SuperCognit Team ### Questions? If anything from your Withnodes setup looks off after the transition, write to hello@supercognit.com and a human will sort it out. New here and curious what the fuss is about? Start with the 7-day free trial — bring a prompt, leave with a product. ## Your website is already an MCP server (you just haven't published it yet) https://supercognit.com/blog/your-website-is-already-an-mcp-server · 2026-07-14 Everything an MCP server for your company would serve is already on your website. Publishing it as an MCP server is a two-minute import, not an engineering project — here is what changes when you do. The answers already exist. Your opening hours, your pricing, your returns policy, what the product does and who it is for — if your website is any good, everything an MCP server for your company would serve is sitting in your pages right now. What is missing is not content. It is a publishing step: making those pages something AI clients can query directly and cite, instead of something they half-remember from a months-old crawl. That publishing step used to be an engineering project. Now it is an import. Paste your URL into SuperCognit and the site is crawled into a cited knowledge base in about two minutes; publish, and you have an MCP server that works in Claude and any other MCP client. The content took you years to write. The server takes an afternoon, generously. ### Machines already read your website — badly Your site has been read by machines for years: search crawlers on their own schedule, and language models through training snapshots of unknown age. So when a customer asks an AI assistant about your cancellation terms, the assistant answers from whatever it absorbed, whenever it absorbed it. Sometimes that is right. You have no way to know, and no way to fix it when it is not. An MCP server inverts this. Instead of hoping models remembered you correctly, you give their clients a live endpoint that serves your current pages, with citations. Same content, different distribution. | | Website as pages | Website as an MCP server | | --- | --- | --- | | Who reads it | People with browsers; crawlers on their own schedule | AI clients, live, on your customers' behalf | | Freshness | Whatever a training snapshot caught | Re-synced on your schedule, up to hourly | | Answer quality | The visitor hunts across pages | Direct answers, each citing its source page | | Your control | None, once crawled | Private, public, or paid per seat | ### Citations are the actual feature The import does not scrape text into a pile. Every fact in the knowledge base keeps its link to the page it came from, so answers arrive with sources attached. That earns trust in both directions: a customer can verify the answer against your site, and you can audit — when an answer is wrong, the citation tells you exactly which page to fix. A knowledge base without citations is a rumor mill with better formatting. ### One import, two products The build that becomes your MCP server is the same build that becomes a chat agent. Put the agent on your website for visitors who want to ask instead of browse; publish the MCP server for the people whose first question goes to Claude rather than to a search box. You maintain one knowledge base and both surfaces stay in sync by construction. It is the same build Four Seasons Fairways runs on. ### Fresh beats big The scheduled re-sync is the quiet advantage. Website sources refresh as often as every fifteen minutes, so the MCP server tracks the site rather than remembering it. A boutique hotel that changes its seasonal rates on Tuesday morning has an MCP server quoting the new rates by Tuesday morning plus fifteen minutes — while models trained on old snapshots keep quoting the previous ones for months. ### Private, public, or a product of its own Publishing does not have to mean public. A private server is an internal company brain only your team can reach. A public one is open to anyone with an MCP client. And if what your site knows is genuinely valuable — specialist know-how, curated reference material — you can sell access per seat: Stripe subscriptions, invoices and payouts are built in, with SuperCognit as merchant of record taking a platform fee. Custom domains are supported, so the endpoint carries your name, not ours. > You already did the hard part. You wrote the website. The gap between companies right now is not who has good content — plenty do. It is whose content AI clients can actually reach. Your website cleared the first bar years ago. The second one is a two-minute import and a publish button. ## How to build your first MCP server for your company — no code required https://supercognit.com/blog/build-your-first-company-mcp-server-no-code · 2026-07-10 Build your first MCP server for your company without writing code: import your website into a cited knowledge base in about two minutes, upload documents, attach tools, publish. Here is the whole process. Building an MCP server for your company no longer requires writing code. The no-code version of the job looks like this: import your website, which is crawled into a cited knowledge base in about two minutes, or upload your documents, attach any tools you want to expose, and publish. What comes out is a working MCP server your team or your customers can connect to from Claude or any other MCP client — no repository, no hosting, no authentication code. That last clause is the one that matters. The Model Context Protocol itself is not exotic, but a production MCP server has always meant four real engineering jobs: write the server, host it, implement auth correctly, and keep the content behind it current. Those four jobs are the reason most companies that would clearly benefit from an MCP server still do not have one. ### What you are actually building An MCP server is an endpoint AI clients can query. Connect yours in Claude and Claude can search your company's knowledge and call your tools mid-conversation, citing the sources it used. The difference is between a model half-remembering your website from an old training crawl and a live, authoritative line to what your company actually says. On SuperCognit, one build produces two surfaces at once: a chat agent you can put in front of customers, and an MCP server that other AI clients connect to. One knowledge base, one set of tools, both surfaces reading from the same source. ### The build, step by step 1. **Import your website** — Paste your URL. The crawler walks your public pages and turns them into a knowledge base where every fact keeps a citation back to the page it came from. For a typical site this takes about two minutes. 2. **Upload documents** — Add what the website does not say: SOPs, internal policies, price lists, product specs. Uploads join the same knowledge base and are cited the same way. 3. **Attach tools** — Optional, and the point where a server becomes more than a reference library. Any HTTP tool you define for your agent — an availability check, a lead capture, an order lookup — becomes a callable tool on the published MCP server. 4. **Choose private or public** — Private servers are an internal company brain: your team connects, nobody else can. Public servers are open to anyone with an MCP client, and you can charge for access per seat if the knowledge is worth paying for. 5. **Publish** — You get an MCP endpoint with authentication already handled. Connect it in Claude and ask it something you know the answer to. Check the citation. ### The authentication you did not have to write Auth is where hand-rolled MCP servers usually go wrong, partly because the spec has kept moving and partly because getting it quietly wrong is easy. Servers published here implement OAuth 2.1 as the current MCP spec describes it — issuer validation, CIMD client registration — with strict per-workspace isolation. Your server authenticates users correctly on day one and you never touch any of it. ### Keeping it current A knowledge base that drifts out of date is worse than none, because it answers confidently and wrong. Website sources re-sync on a schedule you choose, as often as every fifteen minutes, so a price change or a policy edit on your site shows up in MCP answers without anyone re-importing anything. ### Three ways this plays out - A boutique hotel imports its site, uploads the rate sheet and cancellation policy, and publishes. Guests' AI assistants now answer questions about the hotel from the hotel, with citations. - A 40-person accounting firm uploads internal procedures and filing calendars, publishes private, and the whole team gets one authoritative source inside Claude instead of asking the two partners who know everything. - A software vendor imports its documentation site, attaches a license-lookup tool, and publishes public — support questions get answered inside whatever MCP client the customer already uses. Plans are priced in EUR and there is a seven-day trial, which is more time than you need: the import takes two minutes, and the first cited, correct answer from your own company's MCP server takes maybe five more. That answer is the moment this stops being abstract. ## MCP server vs. chatbot: which one does your company actually need? https://supercognit.com/blog/mcp-server-vs-chatbot · 2026-07-06 MCP server vs. chatbot comes down to where your audience is standing: a chatbot meets visitors on your website; an MCP server meets your team and customers inside Claude. How to pick — and why one build can give you both. The short answer: a chatbot serves people who come to your website; an MCP server serves people who live inside an AI assistant like Claude. If your audience is visitors with questions, you need the chatbot. If your audience is your own team, or customers who already work in an assistant all day, you need the MCP server. And since one build can produce both from the same knowledge, the decision is less dramatic than the versus in the title suggests — but the two jobs are worth separating, because they fail differently. ### Two surfaces, one difference that matters Both answer questions from your content. The difference is who owns the window. With a chatbot you own everything: the branding, the welcome message, the domain it lives on. With an MCP server the assistant owns the window, and you supply the knowledge and tools it draws on. Owning the window is a feature when the audience is a stranger on your pricing page. It is a burden when the audience is your own operations team, who already have Claude open and will not visit an internal portal, however nice you make it. ### Side by side | | Chatbot | MCP server | | --- | --- | --- | | Where it lives | Your website or a hosted page | Inside Claude and any MCP client | | Who it serves best | Visitors and customers who come to you | Teams and users who already work in an assistant | | Who controls the experience | You — branding, welcome message, domain | The assistant; you supply the knowledge and tools | | How people reach it | They find your site | They connect your server once, then just ask | | Monetization | Supports the site around it | Access can be sold per-seat, with subscriptions and invoicing built in | ### When the chatbot is the right answer Your audience is website visitors: prospects with pre-sales questions, customers who want an answer at two in the morning, people who would otherwise send an email and wait. They are not going to install anything, connect anything, or learn a new tool for you. The conversation should happen where they already are — on the page — in your brand, opening with your welcome message. A visitor has never heard of MCP and never needs to. That is the chatbot's whole advantage: zero setup on their side, full control on yours. ### When the MCP server is the right answer - An internal company brain: SOPs, policies and product facts your team queries from Claude, behind login, without opening another tab. - Customers who spend their day inside an assistant and want your product's facts available there, not on your website. - Knowledge worth charging for: a firm's methodology or reference material sold per-seat, with Stripe subscriptions, invoices and payouts handled and SuperCognit acting as merchant of record. - Actions, not just answers: an HTTP tool you attach becomes a callable MCP tool, so an assistant can do things on your systems instead of only looking things up. ### The choice is smaller than it looks On SuperCognit these are not two projects. You import your website — crawled into a cited knowledge base in about two minutes — upload documents, attach tools, and publish once. That build is simultaneously a chat agent you can put on your site and an MCP server that works in Claude and any MCP client, private or public, on a custom domain if you want one. The knowledge re-syncs on a schedule, as often as hourly, so both surfaces stay current together instead of drifting apart. Which means the real question is not chatbot versus MCP server. It is: who needs your knowledge, and where are they standing? Serve the visitors on your site with the widget. Serve everyone who already lives in an assistant with the server. One source of truth, two doors — and the 7-day trial is long enough to open both and see which one your audience actually walks through. ## MCP server vs. API: what's the difference and when do you need each? https://supercognit.com/blog/mcp-server-vs-api · 2026-07-01 An API is built for developers; an MCP server is built for AI assistants. A practical comparison of MCP server vs. API — what each one does, where they overlap, and when your company needs each. An API is an interface for software: developers read the documentation, write code against it, and ship an integration. An MCP server is an interface for AI: assistants like Claude connect to it, discover what it offers, and use it in conversation — with no integration code written per consumer. If the people on the other end are programmers, you need an API. If they are people talking to an AI assistant, you need an MCP server. Plenty of companies will end up with both, often with one wrapped around the other. ### The same shape, a different reader Structurally the two are cousins. Both sit on a network, accept requests and return answers. The difference is who is on the other end and what they need in order to succeed. An API assumes a developer with time: they read the docs, handle the errors, and update their code when you change something. An MCP server assumes an AI with none of that: every tool describes itself — name, purpose, inputs — in a form the assistant reads the moment it connects. The documentation is the interface. ### Side by side | | API | MCP server | | --- | --- | --- | | Consumed by | Developers writing code | AI assistants and agents, driven by their users | | Integration cost | Custom code per consumer | Any MCP client connects; nothing to build per consumer | | How capabilities are discovered | A human reads the docs | Tools describe themselves to the assistant at connection time | | Typical contents | Data and operations | Tools plus knowledge — pages and documents the assistant can cite | | Authentication | API keys, assorted OAuth flavours | OAuth 2.1, as the current MCP spec requires | | When you change it | Every consumer updates their code | Assistants read the new descriptions on the next connection | ### When you need an API When other software must call yours predictably, at volume, under a strict contract: a mobile app hitting your backend, a partner syncing orders nightly, a billing job that runs at midnight. APIs are for machines executing plans a human wrote in advance. Nothing about MCP replaces that, and an assistant improvising against your billing system is nobody's idea of progress. ### When you need an MCP server When the consumer is a person working through an assistant. Your team asking Claude about internal policy. A customer asking about your product from inside their own tools rather than on your website. A partner's agent that should be able to check availability or open a ticket without a six-week integration project on their side. In those cases an API technically suffices and practically fails, because nobody on the other end is going to write code — the whole point of asking an assistant is that you didn't have to. ### You probably don't have to choose The neat part is that the two compose. On SuperCognit, an HTTP tool you attach to an agent becomes a callable MCP tool — which means an API you already run can be surfaced through an MCP server without new engineering. The server adds the parts an API lacks for this audience: self-describing tools, cited knowledge sitting alongside the actions, and OAuth 2.1 with per-workspace isolation handled for you. Standing the server up doesn't need engineers either. Import your website — crawled into a cited knowledge base in about two minutes — upload documents, attach the tools, publish. The same build produces a chat agent for your site and an MCP server that works in Claude and any MCP client, private or public, on a custom domain if you want one. If the access is worth money, you can sell it per-seat, with subscriptions and invoicing built in. ### A rule of thumb Ask who has to succeed on the other end. If the answer is a developer with a deadline, build the API and write good docs. If the answer is an assistant with a user behind it, publish the MCP server. And if you already have the API, the MCP server is mostly a wrapper and an afternoon — which is a pleasant amount of effort for a second front door. ## MCP explained for non-developers: why your company knowledge should be one https://supercognit.com/blog/mcp-explained-for-non-developers · 2026-06-26 MCP — the Model Context Protocol — is the open standard that lets AI assistants read your company's knowledge instead of guessing. A non-developer's guide to what it is and why your knowledge belongs on an MCP server. MCP stands for Model Context Protocol. It is an open standard that lets AI assistants plug into outside systems — your documents, your software, your data — roughly the way any browser can open any website. If your company publishes an MCP server, assistants like Claude answer questions from your actual knowledge instead of their training data. That is the entire concept, and no part of it requires you to write code. The name does a lot of unhelpful work, so here is each word in plain terms. Model: the AI. Context: the information the AI has in front of it at the moment it answers. Protocol: an agreed way for two programs to talk. MCP is an agreed way to put your information in front of the AI at the moment it answers. ### The thirty-second technical bit There are two sides. The MCP client is the assistant — Claude, or any other app that supports the standard. The MCP server is your side: it holds knowledge and tools and answers the client's requests. The current spec requires proper authentication — OAuth 2.1 — so a private server is genuinely private. That is all the vocabulary you need for the rest of this article. ### Why your company knowledge should be one Most companies already have the knowledge. It is spread across a website, a policies folder, a handbook nobody has opened since onboarding, and the heads of two long-tenured employees. When someone asks an AI assistant a question about your company today, none of that is in the room. Putting it on an MCP server changes four things at once: - One canonical copy. The assistant reads from the same source every time, instead of whichever stale fragment someone pasted into a chat three weeks ago. - Citations. Answers point back to the page or document they came from, so a wrong answer is traceable and a right one is checkable. - It stays current. Knowledge re-syncs on a schedule — as often as hourly — so the price change you shipped this morning is in this afternoon's answers. - It is permissioned. Internal knowledge stays behind login; public knowledge is deliberately public. The line between them is yours to draw. ### What this looks like on a Tuesday A boutique hotel publishes a private MCP server holding its SOPs, supplier contacts and rate rules. The duty manager asks Claude how to handle a late-checkout dispute and gets the actual policy, cited, instead of a plausible paraphrase of it. A 40-person accounting firm does the same with its engagement templates and filing calendars; the answer to 'what is our position on this' stops depending on who is at lunch. Nothing about either example is futuristic. The knowledge existed already. What changed is that the assistant can reach it. ### The part you genuinely don't have to do You do not have to build the server. On SuperCognit you import your website — crawled into a cited knowledge base in about two minutes — upload whatever documents the site doesn't cover, attach tools if the assistant should be able to do things rather than just know things, and publish. The security requirements the current spec imposes, issuer validation and client registration included, are handled for you, and every workspace is isolated from every other. The same build also produces a chat agent, so the knowledge you organised once serves both the people who visit your website and the people who ask their assistant. And if the knowledge is valuable enough to charge for, per-seat subscriptions, invoices and payouts are built in, with SuperCognit as the merchant of record. ### A reasonable objection 'Our team can just search the shared drive.' They can, and mostly they don't — search returns documents, not answers, and it returns every version of the document with equal confidence. An MCP server is not another place to put knowledge; it is a way to make the knowledge you already have answer questions, wherever the question gets asked. That distinction is why this is worth doing now, while doing it still counts as being early. ## What is an MCP server? A plain-English guide for business owners https://supercognit.com/blog/what-is-an-mcp-server · 2026-06-22 An MCP server gives AI assistants like Claude direct, permissioned access to your company's knowledge and tools. Here's what that means in plain English — and how a business gets one without writing code. An MCP server is a program that gives AI assistants — Claude, and any other app that speaks the Model Context Protocol — direct, permissioned access to your company's knowledge and tools. Instead of an assistant guessing at your prices, policies and procedures from whatever the public internet happens to say, it connects to your MCP server and reads the facts you published, with citations. That is the whole idea. The rest of this guide is what it means for your business. ### Why this exists at all AI assistants are trained on the public internet up to a cutoff date. They have never seen your rate card, your onboarding checklist or the returns policy you updated last Tuesday. So when your staff — or your customers — ask an assistant about your business, it improvises. Sometimes it improvises well. The failure mode is that it improvises confidently. The current workaround is copy-and-paste: someone hunts down the right document, pastes it into the chat, and does it all again next week when the document changes. An MCP server removes that step permanently. The assistant asks your server; your server answers from the source. ### A useful mental model Think of an MCP server as a staffed extension for AI. When an assistant needs something from your company, it calls the extension, asks a precise question, and gets an answer drawn from documents you approved — or presses one of the buttons you chose to expose, like checking availability or creating a ticket. It cannot see anything you didn't publish to it, and your workspace is isolated from everyone else's. ### What lives inside one - Knowledge: your website pages and uploaded documents, indexed so answers come back with citations pointing at the source. - Tools: actions the assistant is allowed to take. On SuperCognit, an HTTP tool you attach to an agent becomes a callable MCP tool automatically. - Access control: MCP servers implement OAuth 2.1 as the current spec requires, so who can connect — and to whose data — is enforced, not assumed. ### What an MCP server is not It is not an API you hire developers to integrate — no one writes code to connect to it; an assistant just does. And it is not a chat widget on your website, although the same build can give you one of those too. It is the layer underneath both: your knowledge and your actions, made available to AI in a form AI can actually use. ### Private brain or public front door The same machinery serves two very different jobs. A private MCP server is an internal company brain: SOPs, policies and product facts, available to your own team inside Claude, behind login. A public one is a front door: anyone with an MCP client can connect and get grounded, cited answers about your business instead of the confident improvisation described above. There is a third option that surprises people: you can sell access. Per-seat subscriptions, invoices and payouts are built in — SuperCognit acts as the merchant of record and takes a platform fee — so a consultancy can package its methodology as a paid MCP server without ever touching Stripe's API. ### Getting one without hiring a developer 1. **Import your website** — Point the platform at your domain. It crawls the pages into a cited knowledge base — for a typical site this takes about two minutes. 2. **Add documents and tools** — Upload the PDFs and internal documents the website doesn't cover, and attach any actions the assistant should be able to take on your behalf. 3. **Publish** — One build produces both a chat agent and an MCP server. Set it private or public, put it on a custom domain if you like, and connect to it from Claude. 4. **Let it track reality** — Knowledge re-syncs on a schedule — as often as hourly — so the server follows your website instead of drifting away from it. ### What it costs to find out Plans are priced in EUR and there is a 7-day trial, which is longer than the import takes by a comfortable margin. The honest test is simple: import your own site, open Claude, connect the server, and ask the three questions your team answers most often. If the answers come back right and cited, you'll understand what an MCP server is in a way no guide can teach you. ## Illusion.ws joins SuperCognit: the AI illusion generator's next chapter https://supercognit.com/blog/illusion-ws-joins-supercognit · 2026-06-17 Illusion.ws — the AI illusion generator behind aesthetic illusion art, hidden-pattern images and optical illusion experiments — has been acquired by SuperCognit. Here's what that means, a practical guide to creating AI illusions today, and what illusion creators can build (and sell) on SuperCognit now. Some news for everyone who found us through AI illusion art: Illusion.ws, the AI illusion generator project, has joined SuperCognit. If you came here searching for illusion AI, aesthetic illusions or an AI illusion generator — welcome. This page explains what happened, and more importantly, what you can create now: both the illusion art you came for, and the AI products that Illusion.ws itself grew into. ### What Illusion.ws was Illusion.ws lived in the most fun corner of generative AI: optical illusion art. The genre that Illusion Diffusion made famous — images where a pattern, a logo, a word or a spiral hides inside an ordinary-looking scene, visible when you squint, step back or let your eyes unfocus. Aesthetic illusions: cityscapes that conceal text, landscapes that resolve into faces, patterns woven invisibly into art. It's the kind of AI output people can't stop sharing, because every image is also a small puzzle. ### Why it joined SuperCognit Building a single-purpose AI tool teaches you a hard lesson: the generator is the easy part. Accounts, billing, hosting, rate limits, languages, support — the platform around the tool is where the real work lives. That's exactly the problem SuperCognit solves: it's the platform where an AI idea becomes a product — with knowledge, tools, domains, channels and billing already built. Folding Illusion.ws into SuperCognit puts the creative-AI know-how where it can reach furthest. ### How to create AI illusions today: a practical guide The craft is very much alive, and the technique is portable across tools (Illusion-Diffusion-style models, ControlNet-based pipelines and their descendants). If you want to make aesthetic illusion art today, this is the method: 1. **Start with a strong control pattern** — The hidden element — a logo, word, spiral or checkerboard — works best as a high-contrast black-and-white image. Bold, simple shapes survive the diffusion process; thin lines and fine detail dissolve. 2. **Pick a scene that can absorb it** — Busy, organic scenes hide patterns best: forests, cityscapes, crowds, clouds, food. Flat minimal scenes leave the pattern nowhere to hide. 3. **Describe the scene, not the illusion** — Your prompt should read like a normal art prompt ('a misty mountain village at dawn, oil painting'). The control pattern does the hiding — the prompt does the beauty. 4. **Tune the control weight** — The single most important dial. Too high and the pattern dominates (obvious, ugly); too low and it disappears. Sweep the middle range and generate in batches. 5. **Squint-test and iterate** — The classic check: squint, shrink the image, or view it from across the room. The best illusions read as art up close and as the hidden pattern at a distance. 6. **Upscale the winners** — Illusions reward resolution — upscale your best results so the scene detail holds up while the hidden structure stays crisp. ### From illusion maker to AI product builder Here's the bigger opportunity. If you loved making AI illusions, you already have the two skills that matter on SuperCognit: prompt craft and taste. The platform turns those into products you can publish and charge for: - An AI art-direction agent — trained on your techniques and style guides, advising users on prompts, control weights and composition, with citations back to your material. - A prompt-engineering coach — package what you know about coaxing models into a subscription agent anyone can ask. - A creative studio assistant — connect image-generation APIs through SuperCognit's HTTP tools so your agent doesn't just advise, it generates. - A niche expert for any craft — the illusion workflow generalizes: whatever you're expert at, an agent can teach it, in 12 languages, around the clock. | | A standalone generator site | Your AI product on SuperCognit | | --- | --- | --- | | What users get | One tool, one trick | An agent with your knowledge, tools and taste | | Revenue | Ads or nothing | Subscriptions — checkout, seats and payouts built in | | Where it lives | One website | Your domain, an embeddable widget, WhatsApp, Telegram, MCP | | Languages | Usually one | 12 out of the box | | Infrastructure | Yours to build and pay for | Hosting, metering, billing and cost caps handled | | Time to launch | Months of engineering | An afternoon — describe it, attach knowledge, publish | > Illusion art taught a generation of creators to prompt. The next step is turning that craft into products — that's the platform Illusion.ws joined. > > — SuperCognit Team ### Start creating SuperCognit is free for 7 days — enough to build your first agent, attach your knowledge, and publish it on your own link. If you're an Illusion.ws user with questions about the transition, write to hello@supercognit.com. And if you just came for the illusions: the guide above is yours — go make something people will squint at.