Docs / Getting started
Your first check
Run a check on any site in about fifteen seconds, without signing up.
Put an address into the checker and wait. No account, no card, no email.
The check fetches what an agent would fetch, from outside, exactly as an agent would. It follows no redirects on the UCP profile, because the specification forbids them there and an agent would not follow one either. Following it would hide the exact failure worth finding.
Reading the result
You get one of four outcomes per standard, and the difference between them matters:
A grade. We found an integration and assessed it. See reading a report for what caps a grade.
Not implemented. Nothing was served at /.well-known/ucp, or nothing WebMCP-shaped appears in
your page. This is a legitimate state, not a failure — plenty of sites have no reason to implement
either standard.
Could not tell. The check did not complete: your origin rate limited us, timed out, or returned a server error. This says nothing about your integration, and we will not pretend otherwise by grading it. Try again shortly.
A blocking failure without an integration. Rare and worth acting on. It means something is actively broken — a profile that redirects, say — found before we could rule the integration in or out.
The WebMCP result is weaker, and says so
WebMCP tools are registered while your page runs, so the only way to be certain what an assistant receives is to evaluate the page in a real browser. The check reads your HTML and your same-origin scripts instead.
That will miss tools added conditionally, and it cannot confirm that registration succeeded. Every WebMCP report says this on its face rather than burying it. Treat it as a strong hint, not proof.
Running it yourself
The same checks are open source and run from your terminal:
npx agentcheck yoursite.com
Same checks, same grades — the CLI and the hosted product share their scoring so the two cannot disagree. See the CLI.
What a check cannot tell you
Whether it still passes tomorrow. A check is a snapshot, and the whole problem with agent readiness is that it breaks quietly between the times you think to look. That is what adding a site is for.