Exposes three tools that help LLMs interpret alerts from a RAG drift monitoring system. When your retrieval pipeline drifts across dimensions like coverage, rank correlation, or embedding shift, interpret_drift_score translates raw metrics into plain English explanations. recommend_thresholds suggests sensible alerting boundaries based on your pipeline's characteristics, and explain_drift_dimensions breaks down what each of the five drift axes actually measures. Part of a fourteen server collection from the mcp-stack monorepo, this one wraps logic from the ragdrift Rust crate. Install it if you're building agents that triage production RAG quality issues or need to turn numerical drift scores into actionable diagnosis without hallucinating what the metrics mean.
claude mcp add --transport stdio io.github.mukundakatta-ragdrift-mcp -- npx -y @mukundakatta/ragdrift-mcp