claude_telemetry

claude_telemetry adds full observability to Claude Code by wrapping the CLI with OpenTelemetry tracing. Every agent run captures tool calls, token counts, costs, execution time, and errors—visible in Logfire, Sentry, Honeycomb, Datadog, or any OTEL backend. The wrapper is a drop-in replacement: use 'claudia' instead of 'claude code' on the command line or call run_agent_with_telemetry() in Python scripts. This solves the visibility gap when running agents headless in CI/CD pipelines, cron jobs, or production environments where console output isn't available. Engineers can debug remote failures without local reproduction, track costs to optimize expensive workflows, and understand exactly what happened during each execution.

Install with pip install claude_telemetry, then use 'claudia' instead of 'claude code' on the command line—all existing flags work unchanged. For Python scripts, import run_agent_with_telemetry and pass your prompt. Configure your observability backend via environment variables (LOGFIRE_TOKEN, SENTRY_DSN, or OTEL_EXPORTER_OTLP_ENDPOINT).