# Machine Authentication & Access Policy — webmcp-tool.com

> webmcp-tool.com exposes public read-only inspection endpoints for AI agents, MCP clients, and developer tooling without requiring API keys or OAuth tokens. Actions that send recurring mail require human double opt-in by email.

Last reviewed: 2026-11-13.

## 1. Unauthenticated Public Endpoints (No API Key Required)

The following machine interfaces are open to anonymous HTTPS clients:

- **Remote MCP Server (`POST https://webmcp-tool.com/api/mcp`)** — Streamable HTTP transport (JSON-RPC 2.0). No `Authorization` header is required for read-only tools:
  - `score_site` (`readOnlyHint: true`): crawl and score any public `http(s)` URL across 25 checks.
  - `explain_check` (`readOnlyHint: true`): explain any individual check's weight, rationale, and fix snippet.
  - `list_checks` (`readOnlyHint: true`): inspect the full scoring ruleset, weights, bonus flags, and conditions.
  - `get_badge_embed` (`readOnlyHint: true`): fetch live Agent Readiness badge markup for eligible domains.
- **HTTP Scan API (`GET https://webmcp-tool.com/api/scan?url=<https-url>`)** — described in [OpenAPI 3.1](https://webmcp-tool.com/openapi.json). No bearer token required.
- **Discovery & Reference Documents** — public GET without authentication:
  - [`/.well-known/api-catalog`](https://webmcp-tool.com/.well-known/api-catalog) (RFC 9727 Linkset)
  - [`/.well-known/mcp.json`](https://webmcp-tool.com/.well-known/mcp.json) (MCP Server Card)
  - [`/.well-known/agent-card.json`](https://webmcp-tool.com/.well-known/agent-card.json) (A2A Agent Card)
  - [`/openapi.json`](https://webmcp-tool.com/openapi.json) (OpenAPI 3.1 specification)
  - [`/llms.txt`](https://webmcp-tool.com/llms.txt) (LLM index)
  - `Accept: text/markdown` content negotiation on public documentation and methodology URLs.

## 2. Stateful / Write Operations & Human Confirmation

- **`watch_site` MCP tool (`readOnlyHint: false`)**: registers a 30-day score-drift watch for an email address. Calling the tool never activates recurring mail directly — it sends a single confirmation email containing a human-operated opt-in link. Do not invoke `watch_site` without explicit user instruction and the user's own email address.
- **Full 25-check report delivery (`POST https://webmcp-tool.com/api/unlock`)**: intended for interactive browser sessions initiated by a user requesting their report by email (capped at 5 reports per company domain per day unless allowlisted).
- **Browser WebMCP tools (`document.modelContext`)**: run inside the visitor's open browser tab using same-origin session state; no external credentials are issued or accepted.

## 3. Rate Limits & Fair Use

- Anonymous scans are rate-limited per client IP to protect target origins and scanner capacity.
- Respect HTTP `429 Too Many Requests` and `Retry-After` headers rather than retrying in tight loops.
- Only public `http` / `https` hosts that resolve to routable public IP addresses are scanned; private, loopback, and link-local ranges are rejected by design.
- Send an honest `User-Agent` header identifying your agent or integration. Usage permissions are stated in [`/robots.txt`](https://webmcp-tool.com/robots.txt) (`Content-Signal: search=yes, ai-input=yes, ai-train=yes`).

## 4. Higher Quotas & B2B / Agency Access

Agencies, CI pipelines, and enterprise teams auditing larger domain portfolios can request raised domain limits or dedicated advisory support via [`https://webmcp-tool.com/contact`](https://webmcp-tool.com/contact) or `mailto:hi@webmcp-tool.com`.
