Your docs are now an MCP server
The reader of documentation is increasingly a coding agent, not a person browsing a page. So every published Git2Docs product is now something an agent can connect to and query directly — an MCP server, an llms.txt index, and clean Markdown on any page. Nothing to configure; it's already live for anything you've published.
Why this matters more for us than for anyone else
Plenty of docs tools now expose an MCP server. The difference is what comes through it. A hand-authored docs site can serve an agent prose that drifted from the product two releases ago — and the agent can't tell. Git2Docs docs are generated from your code and regenerated on every release, and the agent surface serves only sections that passed validation. That makes them the most trustworthy thing you can hand an agent: current by construction, and filtered to what we'd stand behind.
What shipped
- An MCP server per product. One endpoint exposing three read tools:
search_docs(hybrid keyword + semantic search),get_page(a page's full Markdown), andlist_pages(the table of contents). llms.txtandllms-full.txt. The industry-standard index of your product, and the whole corpus in one file — always computed from the live published version.- Markdown on any page. Append
.mdto any documentation URL for a clean Markdown copy — no nav chrome, far fewer tokens than HTML.
How to connect it
Open any of your published docs and look in the footer for “Connect these docs to your AI tools.” It shows the exact MCP URL for that product and a copy button — that panel is the fastest path. The URL follows this shape:
https://<your-docs-domain>/api/mcp/<org>/<product>Then add it to your agent of choice:
- Cursor — add it to
.cursor/mcp.json:{ "mcpServers": { "my-product-docs": { "url": "https://<your-docs-domain>/api/mcp/<org>/<product>" } } } - Claude Code — one command:
claude mcp add --transport http my-product-docs \ https://<your-docs-domain>/api/mcp/<org>/<product> - ChatGPT / other clients — add it as a custom connector (Streamable HTTP) using the same URL. No key required — it serves only what's already public on your docs site.
Prefer plain files? Point any tool at https://<your-docs-domain>/<product>/llms.txt for the index, or append .md to a page URL for its raw Markdown.
See how much of your traffic is already agents
Each product's Configure → Docs for AI tools now reports what share of the last 30 days' docs traffic came from AI agents versus human page views, with a per-surface breakdown. If you'd rather not advertise the connect panel publicly, one toggle there hides it — the endpoints stay available either way.
Nothing to enable
This is on for every published product today. Members fold into their anchor, unpublished repos aren't exposed, and versions can be pinned with a ?version tag — the same rules as your public docs. If your docs are live, your MCP server is too.