Use the scanner from your agent

Everything the report page can tell you, an agent can ask for directly. The scanner runs as a public MCP server at https://webmcp-tool.com/api/mcp over Streamable HTTP. There is no key to request and no account to make; the limits below are the whole contract.

The five tools

Four of the five are read-only and annotated as such. watch_site writes one thing — a pending subscription — and sends one email, and its annotations say so. Nothing is armed until a person confirms.

What it costs you

Two budgets, both per calling address. Reading a cached report, listing checks or explaining one costs a map lookup and is limited to 60 requests a minute. A scan of an address nobody has scored in the last hour makes this machine fetch a stranger's server, so new addresses are limited to 15 crawls an hour. A second call for the same site inside the hour reaches nobody and costs nothing against that budget.

Claude Code

claude mcp add --transport http webmcp-tool https://webmcp-tool.com/api/mcp

{
  "mcpServers": {
    "webmcp-tool": {
      "type": "http",
      "url": "https://webmcp-tool.com/api/mcp"
    }
  }
}

Claude on the web and the desktop app

Free, Pro and Max: Customize → Connectors → Add custom connector. Paste https://webmcp-tool.com/api/mcp, set Authentication to None, add. Team and Enterprise: an admin does the same under Organization settings → Connectors → Add → Custom → Web. The desktop app's claude_desktop_config.json only knows local stdio servers; use the Connectors screen there too.

Cursor

{
  "mcpServers": {
    "webmcp-tool": {
      "url": "https://webmcp-tool.com/api/mcp"
    }
  }
}

Or in one click: Add to Cursor.

VS Code

{
  "servers": {
    "webmcp-tool": {
      "type": "http",
      "url": "https://webmcp-tool.com/api/mcp"
    }
  }
}

ChatGPT

Settings → Security & login → Developer mode, then add the server by URL. No OAuth is required for a server without authentication. Developer mode is available on Pro, Plus, Business, Enterprise and Education plans on the web; check OpenAI's current page for which of those get write access to tools.

From the Claude API

{
  "model": "claude-opus-5",
  "max_tokens": 4096,
  "messages": [{ "role": "user", "content": "Is example.com ready for AI agents?" }],
  "mcp_servers": [
    { "type": "url", "url": "https://webmcp-tool.com/api/mcp", "name": "webmcp-tool" }
  ],
  "tools": [{ "type": "mcp_toolset", "mcp_server_name": "webmcp-tool" }]
}

Anywhere else

Any client that speaks Streamable HTTP and accepts a server without authentication works with the same URL. The server card at /.well-known/mcp.json is the machine-readable version of this page, and its name for the MCP Registry is com.webmcp-tool/agent-readiness. It is also listed on Smithery, which can install it into most clients in one step.

What we record

Every tool call is logged with its tool name, how long it took, whether it succeeded, the shape of its arguments — key names and value types, never values — and the calling user agent. Nothing is tied to a person, and the log is dropped after thirty days. It exists so we can see whether the tools we ask other sites to publish are actually used on this one. The privacy page has the rest.