AI readiness for developer tools and open source

Your docs are the product, and a coding agent is now their heaviest reader.

C1C2P3E3D2E4

The moment

Sam, maintainer of a widely used CLI

A developer asks a coding agent to add the library to a project, and the agent writes code against an API that was removed two majors ago.

What the agent needs: Current install command, current API, current version.

What actually happens: The docs site renders search-driven content client-side, the version selector is JavaScript-only so the agent read an archived page, and there is no llms.txt telling it which docs are canonical. The agent hallucinates the rest.

Which dimensions decide it, and why

AIScan grades five dimensions. These are the ones that carry the outcome for this kind of business.

Content

Docs pages must be server-rendered, versioned in the URL and available as Markdown where possible. A coding agent reading your docs is now a primary user, and it pays token costs for every wrapper element around the text.

Capabilities

An Agent Skills index, an MCP server card or a described API turn documentation into something an agent can use directly. For developer tools this is a genuine adoption advantage, not a compliance exercise.

Discoverability

Canonical URLs per version and a sitemap that lists the current docs prevent the classic failure of an agent confidently quoting version 1 to a version 3 user.

The checks that matter here

Fix guides: /docs/checks/content · /docs/checks/capabilities · /docs/agent-skill

Most sites like this run on Astro — read the Astro playbook for where each fix lives.

Questions we get asked

Why serve Markdown twins of docs pages?

Because the agent reading them pays for every byte of chrome around the content. A .md twin is cheaper to read and less likely to be misparsed. That is check C1.

Should we publish an Agent Skill or MCP server?

If your tool does something an agent could invoke, yes. Checks P2 and P3 cover the discovery files; they are bonus-tier, so they add points without risk.

How do we stop agents citing old versions?

Versioned URLs, a canonical pointing at current, and an llms.txt that names the current docs root. Selector-only versioning is invisible.

Our docs use a search-driven single-page app. Is that fatal?

For agents, largely yes. Pre-render the pages; keep the client search as an enhancement on top.

Does a changelog help?

It is one of the most-read pages by agents deciding whether an answer is current, and it costs nothing you were not already writing.

How do we test this?

Scan the docs root and a deep page. npx aiscan-cli yourdocs.dev --json gives you something you can diff in CI.

Scan your site and see where you actually stand

Free, no account, about twenty seconds. The report names every failing check by ID, shows the evidence we found, and gives the fix for your platform — plus a hand-off prompt you can paste straight into Claude Code or Cursor.

Run a free scan npx aiscan-cli yoursite.com

Prefer to read first? Browse every check we run or the guide library.

Next in Software & agenciesAI readiness for API-first and agent-facing products

Also covers: developer tool, open source project, cli tool, sdk, library docs, api client, framework, package, npm package, python library, documentation site, docs portal, changelog, technical documentation, code examples, agent skill, mcp server.