MODULAR-RAG-MCP-SERVER

Why we included this project

MODULAR-RAG-MCP-SERVER is a working RAG service that exposes its retrieval tools through the Model Context Protocol, so Copilot or Claude can call query_knowledge_hub and related functions directly. The pipeline runs from PDF ingestion and chunking through hybrid BM25-plus-dense retrieval, RRF fusion, and optional reranking, with a Streamlit dashboard for managing collections and tracing queries. Teams looking for a reference implementation of a production-shaped RAG stack rather than a thin wrapper will appreciate that the core components, from the LLM and embedding model to the reranker, splitter, and vector store, are pluggable and swappable by config. It also includes Ragas-based evaluation with golden test sets, so you can measure retrieval quality instead of tuning by feel. The author documents the spec-driven build process and ships Skill-based setup, which makes it a practical learning path for engineers who want to see how a modular RAG system is assembled.