# agents.md — AgentReady

AgentReady scores how ready a website or MCP server is for AI agents. This file tells agents how to use the service directly.

## What you can do

1. **Scan a product**: `POST https://agentready-rho.vercel.app/api/scan` with JSON `{"url": "example.com"}`.
   - `202` → poll the `Location` header URL (`GET /api/scan/status/{id}`) until `analysisStatus` is `complete`, or stream `GET /api/scan/stream?scanId={id}` (SSE).
   - `200` → a fresh-enough cached result (`servedFromCache: true`). Pass `"force": true` to bypass.
   - Optional: `"mcpUrl"` to test a specific MCP server, `"ephemeral": true` for tunnel/localhost targets (never ranked, auto-deleted).
2. **Read a score without scanning**: `GET https://agentready-rho.vercel.app/api/score/{domain}` — 404 means not scanned yet; the body's `nextAction` tells you the exact request to run.
3. **List every check**: `GET https://agentready-rho.vercel.app/api/checks` — stable ids; gate CI on explicit id lists.
4. **Re-verify after a fix**: `POST https://agentready-rho.vercel.app/api/scan/checks` with `{"url", "checkIds": ["access.llms-txt"]}` — always runs live.
5. **Get the fix playbook**: `GET https://agentready-rho.vercel.app/skill.md` — step-by-step instructions to raise a site's score.

## Rate limits

- Scans: 10/minute and 30/day per IP (6/day for `force`). Per-target floors also apply.
- Reads: 60/minute per IP. 429 responses carry `Retry-After`.

## Contact

Report a wrong check result via the "report issue" control on any report page.