Site Passport — MCP server

The protocol layer behind sitepassport.org's live, callable MCP server — an independent, live-verified index of AI-agent readiness for WordPress and other sites.

Live endpoint: https://sitepassport.org/.well-known/mcp.json Fully public, no auth, no API key. GET returns a human/crawler-readable manifest; POST speaks JSON-RPC 2.0 (initialize, tools/list, tools/call) over a single stateless request — no session required.

Connect it

No API key, no signup, nothing to configure — that's the whole pitch.

Claude Code:

claude mcp add --transport http sitepassport https://sitepassport.org/.well-known/mcp.json

Cursor (~/.cursor/mcp.json or a project's .cursor/mcp.json):

{
  "mcpServers": {
    "sitepassport": {
      "url": "https://sitepassport.org/.well-known/mcp.json"
    }
  }
}

VS Code (.vscode/mcp.json):

{
  "servers": {
    "sitepassport": {
      "type": "http",
      "url": "https://sitepassport.org/.well-known/mcp.json"
    }
  }
}

Any other client that supports a remote Streamable HTTP MCP server: point it at the URL above — no headers, no credentials.

What's in this repo, and what isn't

This repo contains the protocol adapter — JSON-RPC request/response handling and the tool schema — the same shape used by other hosted-API MCP servers you may already be pointing agents at (Stripe's agent-toolkit, Perplexity's modelcontextprotocol, Sentry's sentry-mcp): the public repo is the adapter, not the proprietary engine behind it.

server.php in this repo calls sp_normalize_target() and sp_run_checks() — Site Passport's actual scoring engine, which runs against our own private codebase and isn't included here. That's deliberate, for the same reason none of the examples above publish their core business logic: the live endpoint is fully public and callable right now with zero setup, and the exact scoring implementation staying private keeps robots.txt/schema/manifest checks meaningful rather than something a site can trivially game once the precise logic is known.

What you get from this repo:

• The exact MCP protocol shape Site Passport speaks — useful if you're building your own MCP client or debugging a call against the live endpoint.
• The tool schema for check_wordpress_agent_readiness, verbatim.
• METHODOLOGY.md — every check Site Passport runs, its category, and its weight, kept in sync with sitepassport.org/methodology. This is the actual audit surface: what's checked and why, versioned and public, even though the check implementation isn't.

Docker

This repo also ships a Dockerfile for registry checkers (e.g. Glama) that need to start the server in a container and confirm it answers MCP introspection — it is not how you'd actually use Site Passport day to day; use the live HTTP endpoint above for that.

docker build -t sitepassport-mcp .
printf '%s\n%s\n' \
  '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{}}' \
  '{"jsonrpc":"2.0","id":2,"method":"tools/list","params":{}}' \
  | docker run -i --rm sitepassport-mcp

It speaks stdio, newline-delimited JSON-RPC, and answers initialize / tools/list with the real tool schema. tools/call isn't implemented in the container — it needs the private scoring engine described above — and returns a pointer back to the live endpoint instead.

The one tool

check_wordpress_agent_readiness — checks whether a WordPress site (or any website) is ready to be safely operated by AI agents: llms.txt, AI-crawler directives in robots.txt, schema.org markup, and a WebMCP capability manifest. Live-checked against the site itself, never self-reported.

{
  "name": "check_wordpress_agent_readiness",
  "description": "Checks whether a WordPress site (or any website) is ready to be safely operated by AI agents: llms.txt, AI-crawler directives in robots.txt, schema.org markup, and a WebMCP capability manifest. Live-checked against the site itself, never self-reported.",
  "inputSchema": {
    "type": "object",
    "properties": {
      "url": { "type": "string", "description": "The site URL to check, e.g. https://example.com" }
    },
    "required": ["url"]
  }
}

Two of the seven categories in the full methodology — Capability Exposure and Safety & Trust — require the target site to actually expose agent tools behind an approval gate (today, that means running Agent Builder or an equivalent). Those two can't be checked externally, so check_wordpress_agent_readiness covers the five that can: Discoverability, Bot Access, and Content.

Try it

curl -s https://sitepassport.org/.well-known/mcp.json \
  -X POST -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"check_wordpress_agent_readiness","arguments":{"url":"https://wordpress.org"}}}'

Rankings are earned, never bought

A listed site's score is pulled directly from that site's own public manifest, on a recurring schedule — not typed into a form once and trusted forever. See METHODOLOGY.md for the full rationale and sitepassport.org for the live directory.

Support

Questions, issues with the live endpoint, or methodology feedback: hello@agentic-tech.us.

License

MIT — see LICENSE.