Connects Claude to MLflow tracking servers (OSS or Databricks) with 82 tools spanning experiments, runs, model registry, MLflow 3 logged models, traces, assessments, webhooks, and prompt optimization jobs. The big win is aggregation tools like summarize-experiment that fold metric stats and top runs into one response instead of five round trips, plus extractFields projection to cut token overhead. Ships with four workflow prompts (debug failed traces, promote best run, compare top runs) and six resource URIs for inline fetching. Use MLFLOW_TOOLS to scope down to just the categories you need. Defaults to read only mode until you flip MLFLOW_ALLOW_WRITE. If you're doing MLflow 3 workflows or need Databricks trace attachments, this is the only third party MCP that ships them.
claude mcp add --transport stdio io.github.us-all-mlflow -- npx -y @us-all/mlflow-mcp