# AgentReady

> Audit any public URL for agent-readiness — WebMCP tools, structured data, discovery files, forms, and actionability.

AgentReady reads a single HTML page (plus same-origin discovery files) and returns a 0–100 score with actionable fixes. It respects robots.txt, stores nothing, and highlights where AI agents get stuck compared to human visitors.

## Key URLs

- `/` — paste a URL and run an audit
- `/docs` — how the six checks and the scoring work, the honesty rules, and the API, in one page
- `/demo` — live WebMCP demo page: a restaurant booking form with declarative and imperative tools (Le Petit Bistro)
- `/fixtures/soup` — deliberately broken showcase: a page that looks fine to a person and blocks an agent

Grades are not published here: they are produced by the rubric, and the rubric changes. Audit any of these URLs to see its current score.

## API

- `/openapi.json` — OpenAPI 3.1 description of everything below (also linked from `/` as `rel="service-desc"`)
- `POST /api/audit` — body `{"url": "https://example.com/"}` → the full report: `score.overall` (0–100), `score.grade` (A–F), `score.coverage`, and per-category `results` with findings and paste-ready remediation snippets. `GET /api/audit?url=…` is equivalent.
- `GET /api/card?url=…` — the same audit rendered as a 1200×630 PNG share card
- `GET /report?url=…` — the human-readable report, rendered server-side; this is also what the audit form on `/` submits to, and it works with JavaScript disabled

Rate limit: best-effort, ~12 requests/min per client per instance (it is enforced per warm serverless instance, not globally). Nothing is stored. robots.txt is honoured before a page is fetched, so a disallowed URL returns `ok: false` with `code: "robots-disallow"` and no score.

## WebMCP

This site demonstrates declarative (`toolname`, `tooldescription`, `toolparamdescription`) and imperative (`document.modelContext.registerTool`) WebMCP on `/demo`.

`/` registers its own tool: **`audit_site`** — declared on the audit form and registered imperatively from an inline script, taking `{ "url": "…" }` and returning `{ url, overall, grade, coverage, topFinding }`.

Spec: https://webmachinelearning.github.io/webmcp