Agent tools

Your published docs are built to be read by AI agents as well as people.

llms.txt

Every site serves an llms.txt index at /llms.txt (and /.well-known/llms.txt): the site title, its description, your assistant instructions if you have them, and a linked directory of every canonical page grouped by product and section. A full-text dump of the whole site is at /llms-full.txt. Point an agent at /llms.txt and it can discover and fetch everything.

Markdown for every page

Append .md to any page URL to get clean Markdown instead of HTML — /example/guides/quickstart → /example/guides/quickstart.md. The same content is returned to any request that sends Accept: text/markdown. Each Markdown response opens with a short pointer back to /llms.txt so an agent that lands on one page can find the rest.

The page menu

Add a contextual block to docs.json to show a "Copy page" menu beside each page title:

{
  "contextual": {
    "options": ["copy", "view", "assistant", "chatgpt", "claude", "mcp"]
  }
}

MCP server

Every published site hosts a Model Context Protocol server at /api/mcp/mcp so coding agents can search and read your docs directly. It exposes four tools: search_docs, get_page, list_pages, and submit_feedback (which lets an agent report a page that's wrong or outdated). It's on by default; set "mcp": { "enabled": false } in docs.json to turn it off.

Connect it from your agent — Claude Code:

claude mcp add --transport http docs https://your-docs.example.com/api/mcp/mcp

Cursor (.cursor/mcp.json) or VS Code (.vscode/mcp.json):

{ "mcpServers": { "docs": { "url": "https://your-docs.example.com/api/mcp/mcp" } } }