Exposes VAC (Verifiable Agent Credentials) as MCP (Model Context Protocol) tools so AI agents can call the sidecar via MCP.

• vac_request(method, path, body?) — Send an HTTP request through the VAC sidecar. Returns status, response text, and whether a receipt was issued.
• vac_receipts_count() — Info tool (each vac_request uses a fresh client; use the Python SDK for multi-step receipt chains).

1. Install dependencies (from repo root or mcp-server):

   pip install -r mcp-server/requirements.txt
   pip install -e sdks/python   # so vac_client is available

2. Set env (or use defaults):

   • VAC_SIDECAR_URL — default http://localhost:3000
   • VAC_ROOT_BISCUIT — your Root Biscuit token (required for real requests)

3. Run the sidecar (and control-plane, demo-api) so the MCP server can reach it.

stdio (for Claude Desktop / other MCP clients):

cd mcp-server
VAC_ROOT_BISCUIT="<your-token>" python server.py

Or add to your MCP client config (e.g. Claude Desktop) with command pointing to this script and env set.

Streamable HTTP (optional):

If your MCP server supports it, you can switch to mcp.run(transport="streamable-http") and run with a tool like uv run --with mcp server.py.

This component supports listing MCP and Agentic on your skills/resume.