MLflow

Trace, evaluate, improve AI agents with MLflow. Full loop: instrument → trace → evaluate → iterate → validate.

Play video

Instrument, trace, evaluate, and improve AI agents directly from Claude Code using MLflow's open-source observability and evaluation platform. This plugin provides 11 specialized skills that cover the full agent improvement loop — from adding tracing to your code, through systematic evaluation, to targeted issue fixing with regression tests.

Key capabilities include automatic tracing instrumentation for Python and TypeScript across popular frameworks (LangGraph, LangChain, OpenAI, CrewAI, and more), trace analysis and debugging to root-cause agent failures, structured agent evaluation with custom and built-in scorers, and a disciplined fix workflow that writes regression tests before making changes. An intelligent dispatcher skill automatically routes your MLflow requests to the right specialist.

How to use: Ask Claude to work with MLflow naturally — the plugin routes requests to the appropriate skill. Try prompts like: "instrument my agent with MLflow tracing", "evaluate my agent's response quality", "analyze this trace — why did it fail?", "fix this agent issue", or "set up MLflow for my project". You can also use "search MLflow docs for [topic]" to query the official documentation, or "query my MLflow metrics" to fetch aggregated performance data. Requires MLflow 3.8+.