Try cMCP from an existing MCP client¶

Start the cMCP gateway, then configure a client that supports the standard mcpServers stdio shape to launch the included bridge:

{
  "mcpServers": {
    "governed-tools": {
      "command": "cmcp",
      "args": ["client-bridge", "--gateway-url", "https://gateway.example/mcp"],
      "env": {"CMCP_BEARER_TOKEN": "replace-with-the-gateway-token"}
    }
  }
}

Restart the client after changing its configuration. Its tools/list and tools/call messages now pass through cMCP and appear in the gateway audit chain; upstream addresses remain solely in the attested cMCP catalog.

What this proves—and what it does not¶

The bridge is an evaluation aid. It runs on the user's machine, outside the TEE and outside cMCP's measurement. A user who controls the client configuration can remove it, so this does not prove cMCP cannot be bypassed. Production enforcement requires controls that make the governed gateway the only reachable path to upstream servers.

The bearer token is read from CMCP_BEARER_TOKEN; do not put it in command arguments, where process-list tooling may expose it. Diagnostics go to stderr so stdout remains strict newline-delimited JSON-RPC.

Design and trust boundary: #510.