W&B Analytics

MCP server extending Weights & Biases with hyperparameter importance analysis using RandomForest and SHAP, sweep summaries, regression detection, and Plotly chart generation.

Complements the official W&B MCP server with advanced analytical capabilities including hyperparameter importance ranking with R² metrics, sweep summaries combining best and worst run analysis, structured run comparisons, and regression detection using z-tests against baseline runs. Generates Plotly charts exported as inline base64 PNG images. Supports relative-date filtering, optional run and sweep actions via --enable-actions flag, local stdio and HTTP bearer-token transports, and OpenTelemetry observability. Requires Python 3.11+.