Agent readiness checks and scoring
An Agent readiness check grades your website and docs site against 23 scored checks in three stages, plus one check it shows but does not score. The checks come from two open specs, AgentReady v2.0 and the Agent-Friendly Documentation Spec, and a few are EkLine’s own. To run a check, see Check your sites are ready for AI agents.
Levels
Section titled “Levels”Each failing check counts at one of three levels:
| Level | Meaning | Points |
|---|---|---|
| Blocker | Stops AI agents outright, such as a firewall challenge or a page whose text loads only with JavaScript. | 10 |
| Gap | Agents miss or misread something, such as a missing llms.txt or prices shown only in an image. | 3 |
| Polish | Worth doing once the rest passes. | 1 |
Some checks count at a lower level for a smaller problem, such as code blocks without a language tag. The check’s detail under All checks says why it counts as the level it does.
The score
Section titled “The score”The score is the points of the checks that pass, out of the points of every check that applies, as a whole number from 0 to 100. A check that doesn’t apply to you, or that couldn’t run, counts nowhere.
A check shows Not observed when the scan couldn’t see what it looks for, such as an OpenAPI description it couldn’t find or docs pages that came back as a sign-in or challenge page. It doesn’t count against your score. To have The API is described in OpenAPI checked, add the address of your OpenAPI file under Edit sites.
| Band | Score |
|---|---|
| Ready | 80 to 100 |
| Partial | 50 to 79 |
| Blocked | 0 to 49 |
While any Blocker is open, the score stays at 49 or below, whatever else passes.
If the docs site couldn’t be checked, for example because bot protection blocked every request, the score counts the website alone and shows Website only.
Two of your answers about your product change a level:
- An MCP server or CLI agents can act through counts as a Gap until you confirm that AI agents should be able to use your product, and as a Blocker after.
- Docs open without a login, or keep a public index counts as a Gap when you say your docs should need a sign-in.
Can agents reach your pages at all?
| Check | Sites | Counts as |
|---|---|---|
| AI agents allowed in robots.txt, and key pages not hidden by noindex | Website and docs | Blocker |
| No firewall block or challenge for agents | Website and docs | Blocker |
| Docs open without a login, or keep a public index | Docs | Blocker |
| One current page per answer, on your own domain | Docs | Gap |
| Sitemap, listed in robots.txt | Website and docs | Gap |
Can agents read what they reach?
| Check | Sites | Counts as |
|---|---|---|
| The answer is in the first HTML response | Website and docs | Blocker |
| Missing pages return 404, and any 429 carries Retry-After | Website and docs | Gap |
| Homepage links to docs, product and pricing | Website | Gap |
| Prices as text on your own pricing page | Website | Gap |
| Docs formatting: code fenced and tagged, tabs read well as text | Docs | Gap |
| Discovery files are linked from pages agents visit | Website and docs | Gap |
| llms.txt exists, is valid and its links work | Website and docs | Gap |
| Key pages carry JSON-LD; key facts also in text | Website | Gap |
| Pages have markdown versions that match the page | Docs | Gap |
| Pages fit in what an agent reads | Docs | Gap |
| Docs answer through an MCP server | Docs | Polish |
| Docs have a setup or quickstart page, and the homepage links to it | Website and docs | Gap |
Can agents use your product? The checks about an API, MCP server, SDK, or CLI run only for what the check finds on your sites or what you add under Edit sites. These include an OpenAPI file, an MCP server, SDK docs, or CLI docs.
| Check | Shown under | Counts as |
|---|---|---|
| An MCP server or CLI agents can act through | Docs and API | Blocker |
| The API is described in OpenAPI | Docs and API | Gap |
| Agents can get API access on their own | Docs and API | Gap |
| SDKs and a CLI, with install and a first call in the docs | Docs and API | Gap |
| The MCP server answers and its auth can be discovered | Docs and API | Gap |
| Pages where agents take actions register WebMCP tools | Website | Gap |
Stay current
Section titled “Stay current”llms.txt and the sitemap keep up with the docs checks that llms.txt and the sitemap list the pages in your docs navigation and leave out pages that are gone. It shows under All checks but is not scored.
Tracked standards
Section titled “Tracked standards”Under All checks > Tracked, the report also shows whether your sites use newer agent standards. They are not scored:
- AI Catalog
- MCP server card
- MCP Apps
- NLWeb
- agents.json
- A2A agent card
- Content Signals
- Agent payments (x402, MPP, ACP, UCP)
- Cache headers
- Agent skill
Related
Section titled “Related”Built on Potluck Docs, our open source template.