Debugging RabbitMQ, agent-first.
Field notes on diagnosing incidents from your editor — MCP guides, root-cause patterns, and what we're learning building agent-native ops.
Debugging a RabbitMQ consumer crash from your editor
A worker dies, a queue silently fills, and the page comes in. We walk through finding and explaining the whole incident without opening a dashboard — just by asking the agent, which calls Qarote's tools over MCP.
Read the postWiring Claude Code to your broker over MCP
A five-minute setup: mint a key, drop it in .mcp.json, and ask your agent what's wrong.
Why root-cause beats dashboards
Charts tell you something changed. They rarely tell you why. Why explanation, not visualization, is the bottleneck on call.
The four RabbitMQ anti-patterns that page you at 3 a.m.
Missing DLX, idle no-consumer queues, default-exchange overuse, unbounded depth — and how a config scan catches them first.
Reading a depth anomaly: signal vs planned drain
Not every climbing queue is an incident. How the rules engine tells a real backlog from a deliberate drain.
One MCP surface, every client: Cursor, Copilot, Codex
The same Qarote tools from any MCP-capable agent — and what differs between client configs.
Validating licenses offline, with no phone-home
How self-hosted Qarote checks entitlement with signed JWTs so nothing leaves your network — not even a heartbeat.
Debug your next incident by asking.
Connect a broker in under two minutes and wire your agent over MCP.