Agentability
Agentability tests how AI agents use websites and scores 113 sites against agent-readiness checks. Web teams can use its transcripts and site reports to identify barriers and specific fixes.
One of 608 tools in AI Agents
Who it's for
- Web developers maintaining a site who want to find and address agent-readiness check failures.
- Web platform or content teams comparing how well their pages work for agents against other audited sites.
- AI agent researchers reviewing real task attempts and verbatim browsing transcripts.
Not the right fit if…
- Teams seeking an audit of every website: the published ranked panel covers 113 well-known sites.
- Teams needing an agent test that uses JavaScript, logins, forms, or human assistance: the described errands use read-only HTTP requests without those capabilities.
How it fits your workflow
Review a daily episode to see the errands an AI agent attempted, the sites it visited, and whether it finished each task. Transcripts are published verbatim, including failures.
For site-level assessment, browse the index and open a public report. Weekly checks cover llms.txt, crawler policy, readable content, structured data, and MCP; reports list fixes for failed checks. Site owners can request a free audit.
Pricing
Free: The observatory and open data are available at no cost and without signup; the page also describes site audit requests as free.
Public observatory
Free
- · No signup required
- · Open data
- · Free site audit requests
Prices checked on Oct 7, 2026 from the vendor's site. They can change; confirm before you buy.
Key features
- Daily agent errands
- An AI producer turns current search interests into ten real errands, which an AI agent attempts using plain web requests. The site reports results and publishes each transcript verbatim.
- Ranked website index
- A standing panel of 113 well-known websites is scored on agent-readiness signals, with a public ranking and site-level reports.
- Reproducible site checks
- Weekly audits check conventions including llms.txt, crawler policy, readable content, structured data, and MCP. Reports include the exact fix for each failed check.
- Open data and source
- The project describes its data and checks as open and reproducible, with source and data on GitHub and weekly refreshes via public CI.
- Free audit requests
- Site owners can request an audit of any site for free; the page says the audit takes one issue.
Limitations to know
- The agent's errands use read-only HTTP requests and do not use JavaScript, logins, forms, or human help.
- The site's scored panel covers 113 well-known websites; the page does not claim to audit every website automatically.
Getting started
- Open a daily episode and read its published transcript.
- Look up a website in the ranked index and review its score and report.
- Use the report's listed fixes or request a free audit of a site.