Agent readiness

Expose searchable Markdown and structured discovery for AI agents

Docs7 turns each documentation page into clean Markdown for agents and publishes discovery endpoints for the full site.

Agent-readable endpoints

Every Docs7 site publishes these machine-readable surfaces:

The paths below are relative to the documentation root. For a site mounted at example.com/docs, /llms.txt is available at example.com/docs/llms.txt.

Endpoint	Purpose
/llms.txt	Lists routable pages with their Markdown URL and description. Pages in navigation come first.
/llms-full.txt	Combines the complete Markdown content of the site into one file.
/<page>.md	Returns the Markdown representation of one page.
/ask?q=<keywords or question>	Returns the matching pages, headings, and snippets as JSON. Short keywords rank best.
/.well-known/agent-skills/index.json	Advertises a site-specific Agent Skill with links to the docs and the ask endpoint.

An agent can also request a normal page URL with Accept: text/markdown. Docs7 returns Markdown when the client prefers it. HTML responses advertise the Markdown page, both llms.txt files, and the Agent Skills index through Link headers.

If you provide llms.txt or llms-full.txt at the docs root, your file replaces the generated version.

Feedback from agents

Agents can tell you what they could not do with your docs. A hidden note at the top of every page, in HTML and in Markdown, points them to these routes:

Request	Result
GET /feedback?path=…&helpful=…&comment=…	Rates one page. path is the page’s full URL path, base path included, such as /docs/quickstart. helpful is true or false. comment is optional and has at most 500 characters.
GET /requests?title=…&reason=…	Asks for a feature. See Feature requests.

The routes are plain GET URLs, so an agent that can only fetch a page can use them.

Page ratings need "feedback": { "thumbsRating": true } in docs.json. Feature requests need a page with the <FeatureRequests /> component. A site without the setting or the page does not show the routes.

The routes are below the documentation root, as the endpoints above are. A page of your own at /feedback or /requests still opens: Docs7 answers these paths only when the URL has one of the query parameters above. Docs7 refuses these URLs when a browser opens them from another website, so another site cannot add ratings or votes through its visitors.

In browsers that support WebMCP, the page also gives agents the rate_page and request_feature tools for the same actions.

Ratings and requests appear on the AI Visibility tab, in the Feature requests and Page feedback cards. Each card shows how many came from agents and how many from people.

Recover from hallucinated URLs

Agents often request URLs that do not exist. They guess a path from a page title, drop a segment, or reuse a URL from an older version of the docs. Instead of returning an empty 404, Docs7 responds with smart suggestions: the pages the agent was most likely looking for, with their Markdown URLs.

A request for a page that does not exist returns the closest matching pages instead of an empty 404.

For example, an agent that requests /config.md on a site whose page lives at /configuration.md receives that page as a suggestion. The agent can follow it directly without running a separate search, so a wrong guess costs one extra request instead of a dead end.

Write content for agents

Use the same source for people and agents whenever possible. When an agent needs extra context, use Visibility:

<Visibility for="agents">
When the API returns 409, read the existing resource before retrying.
</Visibility>

Content inside for="agents" appears in Markdown output but not on the website. Content inside for="humans" appears on the website and is removed from Markdown output. See the Visibility component.

For reliable retrieval, use exact API names, configuration keys, error messages, and stable headings. Link to the page that owns a fact instead of copying the same instructions into several pages.

Add the site to Context7

Enable Add to Context7 when you connect the repository, or open the Context7 tab for an existing site. A successful production build then refreshes the corresponding Context7 library.

This makes the documentation available through the Context7 MCP server and lets the site’s assistant use the indexed library. Disconnecting stops automatic refreshes but does not remove the library from Context7.