Wiring Claude Code to your broker over MCP

You already live in your editor. Your RabbitMQ broker lives somewhere else, behind a management UI you only open when something is already on fire. Model Context Protocol (MCP) closes that gap: it lets your agent call Qarote's diagnosis tools directly. Setup takes about five minutes.

Mint an API key

In Qarote, open Settings → Agent Access and create a key. It's scoped to the MCP surface — list_incidents, get_incident, explain_incident, and the config-scan tools — and nothing else. Copy it once; you won't see it again.

Add it to .mcp.json

Claude Code reads MCP servers from .mcp.json at your project root. Add a qarote entry pointing at the MCP endpoint, with the key passed as a header:

{
  "mcpServers": {
    "qarote": {
      "type": "http",
      "url": "https://your-qarote.example.com/api/mcp",
      "headers": {
        "Authorization": "Bearer ${QAROTE_API_KEY}"
      }
    }
  }
}

Export the key in your shell so the ${QAROTE_API_KEY} reference resolves:

export QAROTE_API_KEY="qarote_sk_..."

Restart and ask

Restart Claude Code so it picks up the new server, then ask the obvious question:

you ▸ what's wrong with my rabbitmq?

agent ▸ called list_incidents
no open incidents. orders.incoming and events.payment
both draining normally; consumers healthy.

That's the whole loop. When something is wrong, the same question surfaces the open incident and, if you ask, an explain_incident root-cause call.

Detection is deterministic and free — the rules engine flags findings whether or not you ever ask. The agent isn't inventing a diagnosis; it's reading one that already exists, and only spends an LLM call when you explicitly request the explanation.