Developers ask Claude Code, Cursor and Codex which library or API to use, and those agents read your docs directly. For a developer tool, docs that agents can read matter as much as mentions.
Set up
- Prompts. Write them the way a developer types into an agent: “how do i add rate limiting to a next.js api route”, “best postgres hosting for a side project”.
- Engines. Add the main chat engines first, and see Engines for the coding-agent engines (Cursor, OpenCode, Claude Code and Codex), which are available on request.
- Tracker. Install
@usenotra/geo on your docs site as well as your marketing site, and add the docs domain under Tracked domains.
What to work on
- Agent readiness. Serve Markdown to agents that ask for it, since Claude Code sends
Accept: text/markdown (Vercel guide). Then run Agent readiness and fix the Must do items.
- Keep
llms.txt in perspective. Ahrefs found 97% of llms.txt files got no requests and Google does not use it. Coding agents do read it, so publish one, but don’t expect it to move AI search.
- Watch coding agents. The crawler directory lists coding agents such as Claude Code, Cursor and OpenCode, and their visits show which docs pages agents read.
- Follow journeys. Tag Markdown links so you can see the path an agent takes through your docs and where it stops.
- Collect agent feedback. Let agents report what broke with the feedback inbox, and keep in mind that a
feedback.md file also earns an agent readiness bonus.
Metrics to watch
- The Markdown share per source on the Traffic page, which shows how often agents request the Markdown version of a page.
- Cited in answer fetches on docs pages.
- Mention rate on “how do i” prompts.
Last modified on October 8, 2026
