AI agent readiness checker

Can an AI agent use your site?

Free · no sign-in · up to 25 pages

What the agent readiness check covers76 scored checks in 8 groups and 22 suggestions, each with its source.

  • Discover: robots.txt, sitemaps, llms.txt, link headers
  • Access: AI user agents, Markdown negotiation
  • Read and understand: server HTML, language, structured data
  • Act: forms and WebMCP tools agents can use
  • Integrate and pay: API catalog and MCP card, only when relevant

AI agent readiness checker results, explained

Strong (85 to 100)
Agents can find your content, are let in, can read each page as text and can use your forms. Bonus files such as llms.txt or Markdown versions add to this.
Ready with gaps (70 to 84)
Agents get in and can read, but miss helpers: no sitemap or llms.txt for discovery, or form fields without labels.
Important blockers (50 to 69)
Agents hit friction on many pages: bot challenges, pages that need JavaScript for their main text, or buttons without accessible names.
Likely to struggle (under 50)
AI user agents are blocked or served a different page than browsers, so most agent tasks on the site fail before they start.
Only when relevant
API catalog, MCP server card and payment checks apply to sites that offer those things. Elsewhere they are not applicable and do not lower the score.

Common problems with agent readiness

  • Forms agents cannot fill

    Inputs rely on placeholder text instead of a label, and the submit control is a div with a click handler.

    Fix: Give every field a label element, a name and an autocomplete value, and use a real button for submit.

  • Agents treated as attackers

    ChatGPT-User or Claude-User gets a challenge page or a 403 when a person asks an assistant to open your site.

    Fix: Allow verified AI user agents in your CDN or firewall rules, and rate limit instead of blocking outright.

  • Nothing to start from

    No sitemap, no llms.txt and no links in the homepage HTML, so an agent has to guess where the useful pages are.

    Fix: Publish a sitemap and name it in robots.txt, and consider an llms.txt that lists your key pages.

  • Heavy pages for a text reader

    Every page carries large scripts and markup around a few paragraphs, which costs agents time and tokens.

    Fix: Offer a Markdown version through Accept: text/markdown, with Vary: Accept, or at a .md URL.

A sign-up form an agent struggles with, and one it can use

Agents read the accessibility tree, the same names a screen reader announces. Labels and native controls give them something to act on.

FailingStruggles: no labels, no real button
<form>
  <input placeholder="Email">
  <input type="password" placeholder="Password">
  <div class="btn" onclick="signUp()">Go</div>
</form>
PassingUsable: labelled fields, native submit
<form action="/sign-up" method="post">
  <label for="email">Work email</label>
  <input id="email" name="email" type="email" autocomplete="email">
  <label for="pw">Password</label>
  <input id="pw" name="password" type="password" autocomplete="new-password">
  <button type="submit">Create account</button>
</form>

Questions about agent readiness

What is an AI agent in this context?
Software that browses or calls a site for a person, such as ChatGPT agent, Claude with a browser or a coding assistant fetching docs. It reads pages as text and acts on forms and APIs.
Is agent readiness the same as accessibility?
They overlap a lot. Agents rely on the accessible names, labels and roles that assistive technology uses, so fixing accessibility often helps agents too.
Do I need llms.txt, WebMCP or an MCP server?
No. They earn a bonus when present. The scored basics are access, readable HTML, structured data and usable forms.
Why does the checker ask my site for Markdown?
Some sites answer a request with Accept: text/markdown with a Markdown version of the page, which an agent can read with far fewer tokens than HTML. The checker asks for it, checks that the reply is Markdown sent with Vary: Accept so caches keep the two versions apart, and that browsers still get HTML.