An MCP server and an A2A agent, for finding out what AI assistants can actually read.

Three tools. No API key, no signup, nothing to pay. Point it at a URL and it tells you which AI crawlers you block, how much of your page survives without JavaScript, and what a model is given to work with when someone asks about you.

Add it

Claude Code

claude mcp add --transport http agentbuilt https://agentbuilt.dev/api/mcp

Claude.ai and Claude Desktop

Settings → Connectors → Add custom connector, and paste https://agentbuilt.dev/api/mcp. No authentication.

Cursor (.cursor/mcp.json)

{
  "mcpServers": {
    "agentbuilt": { "url": "https://agentbuilt.dev/api/mcp" }
  }
}

VS Code (.vscode/mcp.json)

{
  "servers": {
    "agentbuilt": { "type": "http", "url": "https://agentbuilt.dev/api/mcp" }
  }
}

Anything that speaks JSON-RPC

curl -s https://agentbuilt.dev/api/mcp \
  -H 'content-type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call",
       "params":{"name":"audit_ai_readiness",
                 "arguments":{"url":"example.com"}}}'

Streamable HTTP, stateless, protocol version 2025-06-18. A GET on the same URL returns a plain description of the server.

The tools

What the free audit does not tell you

Readiness is whether assistants can read you. It is not whether they do recommend you. A site can score 100 and still never be named, because being legible is a precondition, not a ranking.

The only way to answer the second question is to ask the models the questions your buyers ask and count what comes back. That is the paid measurement, and the method is published in full at how to check whether ChatGPT recommends your brand — you can run it yourself for the price of the API calls. If you would rather not, we run it.

Agent-to-agent

The same three skills are exposed over the Agent2Agent protocol, so another agent can delegate to agentbuilt as a peer instead of calling it as a tool.

curl -s https://agentbuilt.dev/api/a2a \
  -H 'content-type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"message/send","params":{
        "message":{"role":"user","kind":"message","messageId":"1",
          "parts":[{"kind":"text","text":"Is example.com readable by AI assistants?"}]}}}'

Limits, stated up front

• Only http and https on ports 80 and 443. Hosts resolving to private, loopback, link-local or reserved addresses are refused, so this cannot be used to probe an internal network.
• One page per call, 8-second timeout, at most three redirects, response body capped.
• The robots.txt reading is deliberately simple: it answers "is this agent disallowed from the whole site", not every path-level rule.
• It fetches raw HTML. A site that renders everything client-side will score badly here — which is the point, because that is also what the assistants see.