What is CMD+K?
CMD+K is an in-product chat assistant you embed in your own app, powered by an MCP (Model Context Protocol) server you control and grounded in your own documentation. You drop one script tag, sign a short-lived JWT for the logged-in user from your backend, and the assistant answers questions and takes real actions — read your data, create a ticket, update a record — using whatever tools your MCP exposes.
Who is CMD+K for?
CMD+K is built for product, support, and platform teams that already have an authenticated web app and want to add a real assistant — not just a docs search bar — without writing the LLM plumbing themselves.
- SaaS product teams who want a chat-driven shortcut layer over their app (open a ticket, change a setting, look up a record).
- Developer-tools companies with an API who want an in-product copilot that can read and write through that API.
- Support teams tired of brittle scripted flows and looking for a grounded assistant that cites the docs it answers from.
What are the three pillars of CMD+K?
The product is organized around three layers, each usable on its own:
- Build MCP — the hosting layer for MCP servers. Convert a REST API into an MCP server, or host an existing one behind our gateway with auth, custom domains, and analytics. See Build MCP overview.
- CMD+K Widget — the embeddable chat assistant. One script tag, JWT-authenticated as your end-user, and it pulls in knowledge sources and MCP tools you assign. See Widget overview.
- Monetize — an optional add-on for operators who want to resell AI usage to their end-users — with per-user attribution and Stripe meter forwarding. $99/mo flat. Pricing is on Pricing & plans; full Monetize docs land in phase 2.
MCP, in one sentence
MCP (Model Context Protocol) is an open standard from Anthropic that lets AI assistants call your tools through a single uniform HTTP interface — replacing one-off function-calling integrations.
What do you get out of the box?
- Hosted LLM inference (managed models — no API key to bring).
- Knowledge ingestion (URL, sitemap, file) with citations.
- MCP gateway with end-user OAuth proxy and custom domains.
- Per-end-user identity via short-lived JWTs you mint server-side.
- Built-in usage metering, spending caps, and billing.
- SAML SSO and whitelabel on the AI Pilot plan.
Next steps
FAQ
- Is CMD+K a chatbot builder or an LLM provider?
- Neither, exactly. CMD+K is the runtime that connects an LLM, your knowledge base, and your tools (via MCP) into one chat surface inside your product. We host the inference on managed models: Command+K itself does not support BYOK, so you never bring your own AI keys.
- Do I need an MCP server to use CMD+K?
- No. You can launch a docs-only assistant with just a knowledge source. MCP is what gives the assistant actions (read/write tools). You can add it later without changing the embed.
- How is CMD+K different from a generic chatbot widget?
- Generic widgets either answer from a knowledge base or run scripted flows. CMD+K runs as the signed-in end-user, calls real APIs via MCP with the user's permissions, and cites the docs it read.
- Where does CMD+K run?
- The widget runs in an iframe on your site. All chat, tool calls, and knowledge retrieval happen on our backend (Cloudflare Workers + Postgres). Your jwt_secret never leaves your server.