doc2mcp is live — turn any docs into a Cursor-ready MCP in 60 seconds. Try it free

If you build with AI coding tools, you have hit this wall: the model is confidently wrong about a library because it never read the current docs. doc2mcp fixes that. Paste a documentation URL, and you get a hosted Model Context Protocol (MCP) server that Cursor, Claude, OpenAI Agents, and any MCP client can search, read, and cite.

The problem

LLMs are trained on a snapshot of the internet. The moment a framework ships a new API, your assistant starts hallucinating. The usual workarounds — copy-pasting docs into the chat, or maintaining a custom RAG pipeline — are slow, stale, and a pain to keep alive.

What doc2mcp does

doc2mcp turns a docs site into a live, hosted MCP server in a couple of clicks:

Paste a URL — point it at any documentation site.
We crawl and index it — pages are fetched, chunked, and embedded.
You get an MCP endpoint — secured with a token, ready to plug into any MCP client.

Your assistant can now call an ask_documentation tool that answers questions grounded in the real docs, with source links for every claim.

Why it is different

Hosted, not local. No Docker, no cron jobs, no re-scraping. The MCP stays live and fresh.
Token-based and shareable. Every MCP has its own bearer token. Add it once to your editor and forget it.
Marketplace. Public MCPs show up in the doc2mcp marketplace, so you can reuse what others already built instead of re-indexing the same docs.
Cites its sources. Answers come back with the exact pages they were drawn from.

Use it in your editor

Once you have an MCP URL and token, drop it into your client config. For Cursor (~/.cursor/mcp.json):

{
  "mcpServers": {
    "my-docs": {
      "url": "https://doc2mcp.site/api/mcp/<project-id>/mcp",
      "headers": { "Authorization": "Bearer <your-token>" }
    }
  }
}