Agent-Readiness Auditor scores how well agents like Claude, Gemini and ChatGPT can read your page, find its buttons and finish tasks on it. Then it shows you exactly what to fix.
Runs entirely in your browser. No account, no tracking.
Why another checker?
Agents read your page through its accessibility tree, the same structure screen readers use. A button with no name, a field with no label or a cookie banner in the way can stop them cold, no matter what’s in your robots.txt.
Features
Type what a visitor would ask an agent to do, like “Add this backpack to the cart”. Chrome’s built-in AI reads your page exactly as agents do and plans the clicks, then the auditor shows each step on the page and where an agent would get stuck.
It runs on your computer with Gemini Nano: free, private, no API key. On computers without it, the test matches the task to control names.
Audit up to 25 pages from your sitemap or navigation in one go, free. The scan shows which problems repeat across pages, and a problem on most pages almost always comes from one shared template or component.
Copy one fix prompt for the shared problems, open a site report for your team or client, or download a CSV.
Fix these first picks the three changes worth the most points and shows the score you’d reach. Mark issues on page puts a numbered marker on every problem element at once; hover a number for details.
The Agent view shows the page as an agent reads it: its accessibility tree as text, with a token estimate. Buttons and links with no name are flagged in red, because to an agent they’re just “button”.
Select any line to find it on the page, filter to unnamed controls, or copy the whole snapshot.
On product pages, the auditor checks what shopping agents rely on: Product structured data, price and stock (and whether the price matches the page), shipping, returns and ratings, and an “Add to cart” that’s a real, clearly named button.
Every problem element comes with suggested code you can copy: a real button instead of a clickable div, a label for a field, a name for an icon button guessed from the page. Or copy the whole report, the top fixes, or a single issue as a ready-made task for Claude Code, Cursor or Copilot. It tells the agent to fix each issue at its source, write a test, verify it and keep going until the checklist is done.
Share report makes a clean, standalone report you can send to a client or save as PDF.
Finds hidden text aimed at AI, like “ignore previous instructions” in an off-screen span or an HTML comment.
Run it again after a fix to see your score change and which issues are fixed, new or reduced.
After an audit, the toolbar icon shows the page’s score, so you see it at a glance as you work.
Put your page next to any page you’ve audited: scores by category, and where each one does better.
Editing your site? The audit runs again every time you save and the page reloads.
Corrected markup for each problem element, native HTML first, ready to copy. On-device AI can suggest names for unnamed buttons.
Turn it on after a site scan. The same pages are checked every week, and you get an alert if the score drops.
Fail a CI build below a score with npx agent-readiness-audit, or let Claude Code and Cursor audit pages through MCP.
English, Español, Deutsch, 日本語, Português and বাংলা, following your Chrome language.
What it checks
Each area gets a score, weighted into one overall score from 0 to 100. Checks that don’t apply to the page are left out rather than counted against you.
Clickable divs that aren’t real buttons, links that go nowhere, controls the Tab key skips, no visible focus.
Icon-only buttons, vague names like “Click here”, missing alt text, misleading ARIA.
Unlabeled fields, missing input types and autocomplete hints, unmarked required fields.
Cookie banners and overlays, CAPTCHAs, cross-origin frames, tiny targets, layout shifts. Shadow DOM and same-site frames are checked too.
Controls only reachable by generated class names or auto-generated IDs.
Title, language, landmarks, headings, page size, content that needs JavaScript.
Only on product pages: Product data, price and stock, shipping, returns, ratings, a clear Add to cart.
Hidden text that reads like instructions to an AI agent.
Meta description, Open Graph, structured data, llms.txt, robots.txt rules for AI.
Markdown for agents, MCP server card, API and OAuth discovery, Content Signals, WebMCP.
How it works
Your site, a competitor’s, or a staging build. It works on any regular web page.
Or press Alt+Shift+A. The side panel opens next to the page and the audit starts.
Work through the issues yourself or hand them to a coding agent, then audit again to see what changed.
Private by design
The audit runs inside Chrome on the page you choose. The extension has no servers, so there’s nothing to send your data to.
Read the privacy policyFAQ
Any agent that reads web pages through the accessibility tree or the DOM, including Claude in Chrome, Gemini in Chrome, ChatGPT agent, Perplexity Comet and agents built on Playwright or browser-use. Agents that work from screenshots benefit too: real buttons, clear names and stable layouts help them as well.
There’s a lot of overlap, because agents and screen readers read the same structure, and the extension shows it: checks that are also WCAG 2.2 failures are marked with their success criteria, and a WCAG filter lists them. But it also checks things accessibility tools don’t: hidden prompt injections, content that only appears after JavaScript runs, selectors that break on every release, how many tokens come before your main content, and agent protocols like MCP and WebMCP.
No. Protocol files are shown for information and don’t affect the score. Few sites publish them and few agents read them yet. Whether an agent can actually use your page matters far more.
No. It uses Chrome’s built-in model, Gemini Nano, which runs on your computer. Chrome downloads it once, on the first test, if your computer meets Chrome’s requirements. Without it, the test matches your task to control names inside the extension. Nothing leaves your browser either way. Press Run the steps to watch the plan carried out on the live page in a background tab; it never pays, places orders, deletes anything or types personal details.
Yes, up to 25 pages at a time, for free. Open the Site scan tab, choose pages from your sitemap or your navigation, and the extension audits them one by one in a background tab. It asks for access to that one site only.
Yes. The same checks are available as a command line tool: npx agent-readiness-audit http://localhost:3000 --min-score 80 fails the build when the score drops below 80. Run npx agent-readiness-audit mcp as an MCP server and coding agents like Claude Code or Cursor can audit a page, fix it and audit again on their own.
The extension follows your Chrome language: English, Spanish, German, Japanese, Portuguese (Brazil) and Bengali. Exported reports and prompts for coding agents stay in English.
No. The extension is free, with no account and no usage limits.
The checks are heuristics: they find likely problems fast and point to the exact elements. Use the score to find and prioritize fixes, then confirm with a real agent run on your most important flows.