Skip to main content

Offer a public MCP server

More and more of your users engage with your product through AI: they ask ChatGPT or Claude about it, and coding agents like Claude Code and Cursor work with it on their behalf. Increasingly, they expect you to support that, and most companies now ship an MCP server exposing close to their full API surface. Whatever else your MCP server offers, one of its most important tools is access to the knowledge about your product. A large share of what agents need is not an API call but an answer about how the product works, and without that knowledge they cannot answer product questions, make sense of what your API returns, or know what your product can do in the first place.

Ways to build this with Kapa​

  • Let Kapa host one for you. The hosted MCP server is ready-made: if you do not run an MCP server yet, this is all you need, a public server that lets your users' agents, Claude Code, Codex, Claude, and any other MCP client, search for information in your documentation, community discussions, and code. Kapa hosts it; there is nothing for you to operate. Your users connect it to whichever assistant or editor they work in.
  • Expose Kapa's agentic retrieval inside your own MCP server. If you already ship a server with your platform tools, add documentation search to it, either by proxying the hosted server so users connect to one server and get both, or by wrapping the Retrieval endpoint as a tool you define yourself.

What to index​

Strictly public sources, since the server is offered to the world: your documentation via Website Crawl, open source repositories via the GitHub connectors, and community threads via Discourse, Stack Overflow, or GitHub Discussions. Start with Index your first source.

Resources​