Monitors in Claude Code: Track Performance and Quality in Real Time

What Monitors Do

Monitors are real-time dashboards and logging outputs that show you what Claude Code is consuming and producing as it works. They are especially useful for long-running sessions where you want to catch problems — like a token budget being consumed faster than expected, or an agent going in the wrong direction — before the session ends and you have to start over.

Built-in Monitor

Claude Code's status line is the default monitor. It shows context usage and active tools. Enable verbose mode for more detail:

claude --verbose

This writes detailed logs to claude-debug.log including every tool call, token count, and model response latency.

Custom Monitor Integration

For team deployments, you can pipe Claude Code output to your own monitoring system:

claude --verbose 2>&1 | python3 ./scripts/monitor.py

The monitor script receives structured JSON events and can forward them to Datadog, Prometheus, or whatever observability stack you use.

What to Watch

The most useful metrics to monitor:

• Token consumption rate — catch context saturation early
• Error rate — frequent tool failures usually indicate a permissions or path issue
• Session duration — unusually long sessions often mean the agent is stuck in a loop