🧠 Ragmint MCP Server

AI-Powered Optimization for RAG Pipelines

This server provides 6 MCP Tools for RAG pipeline tuning, dataset generation & workspace control — all programmatically accessible through MCP clients like Claude Desktop, Cursor, VS Code MCP Extension, and more.

🔧 MCP Tools

• 📄 Upload Docs: Upload .txt files to workspace for evaluation using upload_docs.
• 🔗 Upload URLs: Import remote docs via URLs with upload_urls.
• 🔧 Optimize RAG: Full hyperparameter search (Grid/Random/Bayesian) with metrics on optimize_rag.
• ⚡️ Autotune RAG: Automated recommendations for best chunking and embeddings with autotune.
• 🧩 Generate QA Dataset: Create validation QA pairs with LLMs for benchmarking using generate_qa.
• 🗑️ Clear Cache: Reset workspace and delete stored docs with clear_cache.

🧠 What Ragmint Solves

• Automated RAG hyperparameter optimization.
• Retriever, embedding, reranker selection.
• Synthetic validation QA generation.
• Evaluation metrics (faithfulness, latency, etc.).
• Experiment tracking & reproducible pipeline comparison.

🔬 Built for RAG engineers, researchers, and LLM developers who want consistent performance improvement without trial-and-error.

⚙ Powered by

• Optuna (Bayesian Optimization).
• Google Gemini 2.5 Flash Lite/Pro.
• FAISS, Chroma, BM25, scikit-learn retrievers.
• Sentence-Transformers/BGE embeddings.

🌐 MCP Connection

HuggingFace Space
https://huggingface.co/spaces/andyolivers/ragmint-mcp-server

MCP Endpoint (SSE — Recommended)
https://andyolivers-ragmint-mcp-server.hf.space/gradio_api/mcp/sse

📦 Example MCP Use Cases

• Run Auto-Optimization for RAG pipelines.
• Compare embedding + retriever combinations.
• Automatically generate QA validation datasets.
• Rapid experiment iteration inside Claude/Cursor.