@a2a-compliance/mcp

Model Context Protocol server for the A2A (Agent2Agent) protocol compliance test kit. Lets Claude Desktop, Cursor, Codex, Aider, Cline, Windsurf, Continue, and any other MCP-capable client invoke run_compliance / validate_agent_card / list_checks / explain_check / ssrf_check_url as native tools — no REST, no API keys, no shell-out.

Part of a2a-compliance. For CLI use, install @a2a-compliance/cli instead.

Install

The server is published on npm and designed to be launched by an MCP client via npx:

# sanity check
npx @a2a-compliance/mcp
# ^ waits on stdio for an MCP client — Ctrl+C to exit

Configure your MCP client

Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "a2a-compliance": {
      "command": "npx",
      "args": ["-y", "@a2a-compliance/mcp"]
    }
  }
}

Restart Claude Desktop. The tools appear in the tool picker ("⚒" icon) as run_compliance, validate_agent_card, list_checks, explain_check, ssrf_check_url.

Cursor

~/.cursor/mcp.json (global) or .cursor/mcp.json (per-project):

{
  "mcpServers": {
    "a2a-compliance": {
      "command": "npx",
      "args": ["-y", "@a2a-compliance/mcp"]
    }
  }
}

Codex / Cline / Windsurf / Continue

Same shape — invoke via npx @a2a-compliance/mcp with stdio transport. Any MCP client that supports the public spec will work.

Tools exposed

| Name | Input | What it does | |---|---|---| | run_compliance | { url, skipProtocol?, skipSecurity?, skipAuth? } | Full compliance run — agent card + JSON-RPC + method set + auth + security. Returns a ComplianceReport with per-check status and a compliance tier (NON_COMPLIANT / MANDATORY / RECOMMENDED / FULL_FEATURED). | | validate_agent_card | { url } | Card-only, faster. | | list_checks | {} | Returns the full check catalog: id, category, severity, title, description, optional specRef pointing back into the A2A / JSON-RPC spec. | | explain_check | { id } | Returns the catalog entry for one check id. | | ssrf_check_url | { url } | Ingress guard — refuses private-space resolutions. Usable by other MCP tools before they fetch user-supplied URLs. |

Example prompts

• "Check whether https://agent.example.com is a compliant A2A agent and tell me what's wrong."
• "List every A2A compliance check in the security category."
• "Why does sec.ssrf exist? Explain it."
• "Before you fetch this URL, SSRF-check it."

Security

• stdio-only transport — no network listener, no auth surface.
• Every outbound probe uses @a2a-compliance/core's hardened HTTP client: 10 s timeout, 2 MB body cap, per-redirect SSRF re-check, and DNS-rebinding pin (close-the-TOCTOU lookup through undici).
• The ssrf_check_url tool is exposed specifically so an MCP host can sanitise URLs it received from a user before handing them to any fetching tool — including this one.

See also

• 🏠 Repository + full docs
• 🔌 @a2a-compliance/cli — command-line use
• 🧱 @a2a-compliance/core — library form
• 🛡️ A2A Security Top 10
• 📜 Model Context Protocol spec

License

MIT.