Agent Readiness: How GenReady Checks Whether AI Agents Can Use Your Site
The eight agent standards we probe for, how they affect your score, and which one to add first.
What this section of your report checks
Most of your report is about whether AI search engines can read and cite your pages. Agent Readiness asks a different question: can an AI agent do something on your site? Book, buy, query your data, call your API, pay for a resource.
Agents find out what a site can do by fetching a small number of known files and endpoints. We check eight of them on every scan:
| Standard | What we fetch | Who it is for |
|---|---|---|
| Markdown for Agents | Your page, with Accept: text/markdown | Any site with content worth reading |
| MCP Server Card | /.well-known/mcp/server-card.json | Products that run an MCP server |
| A2A Agent Card | /.well-known/agent-card.json | Products that run their own agent |
| Agent Skills | /.well-known/skills/index.json | Products with an API or a workflow to teach |
| webMCP | Inline scripts in your page | Sites with forms, search, or checkout |
| NLWeb | POST /ask | Large, structured catalogs |
| UCP Commerce | /.well-known/ucp | Online stores |
| x402 Payments | /.well-known/x402.json | Paid APIs and paid content |
How it affects your score
Agent Readiness is bonus-only. Each standard we detect adds 1 point to your total score, up to 4 points. A site with none of them loses nothing, which is where nearly every site on the web is today.
Each check ends in one of three states. "Detected" means we found a working implementation. "Not detected" means we looked and it was not there, or it did not match the format. "Could not check" means the request timed out, or your server answered with a rate limit or a 5xx error. That result is inconclusive, and the next scan tries again.
Why a 200 response is not enough
Many sites answer every unknown URL with their homepage and a 200 status. If we counted that, every single-page app would pass every check. So we never trust the status code alone. A file must come back as JSON, not HTML, and must contain the fields an agent needs to use it. Each article in this section lists those fields.
Where to start
For a content site or blog, start with Markdown for Agents. It is often a setting in your CDN or docs platform.
For a SaaS product with an API, publish an Agent Skills index first, then an MCP server card if you already run an MCP server.
For an online store, look at UCP. Most large commerce platforms can generate the manifest for you. If you sell data or API access per request, look at x402.
Publish one standard at a time and rescan after each. They live at separate paths and do not depend on each other.
💡 Quick win
Run curl -sI -H "Accept: text/markdown" https://yoursite.com/. If the content-type line says text/markdown, you already pass one check. If it says text/html, start with the Markdown for Agents article.