This MCP server provides RAG retrieval evaluation metrics for an LLM to compute on demand. It focuses on information retrieval scoring functions including recall@k, hit@k, MRR, and NDCG@k, and includes an evaluate_batch helper to run the metrics across multiple queries.

🛠️ Key Features

• Computes Recall@k, Hit@k, MRR, and NDCG@k
• Binary relevance NDCG@k with log2 discount
• evaluate_batch to evaluate multiple queries in one call
• Exposes tools: recall_at_k(retrieved, relevant, k), hit_at_k(retrieved, relevant, k), mrr(retrieved, relevant), ndcg_at_k(retrieved, relevant, k)

🚀 Use Cases

• Evaluate RAG retrieval performance for ranking quality
• Score retrieved results against relevant items per query
• Apply the same metrics to many queries via evaluate_batch

⚡ Developer Benefits

• Standard metric coverage for IR evaluation (Recall@k, Hit@k, MRR, NDCG@k)
• Clear tool signatures matching retrieved, relevant, and k inputs

⚠️ Limitations

• Metrics are defined with binary relevance for NDCG@k; the readmeExcerpt does not describe graded relevance support.