Easy MCP RAG 🚀

A high-performance Model Context Protocol (MCP) server for RAG using Qdrant. Built for UV/UVX with CPU/GPU support and HTTP transport.

✨ Features

• 🔍 Automatic Document Indexing - Scan directories and index all documents
• 📁 Smart Organization - Each subdirectory becomes its own searchable dataset
• 🛠️ Dynamic MCP Tools - Auto-generated tools for each collection
• 📄 Multi-Format Support - PDF, DOCX, CSV, XLSX, TXT, Markdown, and more
• ⚡ GPU Acceleration - Optional CUDA/MPS support for faster embeddings
• 🌐 HTTP Transport - Run as HTTP server or stdio
• 📦 UV/UVX Ready - Install and run with a single command
• 📊 Verbose Logging - Detailed query tracking and monitoring

🚀 Quick Start

Install with UVX (Recommended)

Run directly from GitHub without installation:

uvx --from git+https://github.com/yourusername/easy_mcp_rag.git easy_mcp_rag --data-dir ./documents

Install with UV

# Install from GitHub
uv pip install git+https://github.com/yourusername/easy_mcp_rag.git

# Or clone and install locally
git clone https://github.com/yourusername/easy_mcp_rag.git
cd easy_mcp_rag
uv pip install -e .

📋 Prerequisites

1. Start Qdrant (using Docker):

docker run -p 6333:6333 qdrant/qdrant

2. Prepare your documents:

documents/
├── legal_docs/
│   ├── contract.pdf
│   └── terms.docx
├── research/
│   ├── paper1.pdf
│   └── notes.txt
└── data/
    └── analysis.csv

💻 Usage

Basic Usage (stdio)

# With UVX
uvx --from git+https://github.com/yourusername/easy_mcp_rag.git easy_mcp_rag --data-dir ./documents

# With UV
uv run easy_mcp_rag --data-dir ./documents

# After installation
easy_mcp_rag --data-dir ./documents

HTTP Mode

easy_mcp_rag --data-dir ./documents --transport http --http-port 8000

GPU Acceleration

# Auto-detect GPU
easy_mcp_rag --data-dir ./documents --device auto

# Force CUDA (NVIDIA GPU)
easy_mcp_rag --data-dir ./documents --device cuda

# Force MPS (Apple Silicon)
easy_mcp_rag --data-dir ./documents --device mps

# Force CPU
easy_mcp_rag --data-dir ./documents --device cpu

Advanced Configuration

easy_mcp_rag \
  --data-dir ./documents \
  --qdrant-host localhost \
  --qdrant-port 6333 \
  --device cuda \
  --embedding-model all-mpnet-base-v2 \
  --chunk-size 1024 \
  --chunk-overlap 100 \
  --top-k 10 \
  --batch-size 64 \
  --verbose \
  --force-reindex

🔧 Configuration Options

Flag	Description	Default
--data-dir	Directory with document subdirectories	Required
--qdrant-host	Qdrant server host	localhost
--qdrant-port	Qdrant server port	6333
--device	Device: auto, cpu, cuda, mps	auto
--transport	Transport type: stdio, http	stdio
--http-host	HTTP server host	0.0.0.0
--http-port	HTTP server port	8000
--embedding-model	Sentence transformer model	all-MiniLM-L6-v2
--chunk-size	Text chunk size (chars)	512
--chunk-overlap	Chunk overlap (chars)	50
--top-k	Results per search	5
--batch-size	Embedding batch size	32
--verbose	Enable verbose logging	False
--log-level	Log level	INFO
--force-reindex	Force reindex all docs	False

🎯 MCP Client Configuration

Claude Desktop / Cline / Other MCP Clients

Add to your MCP client config:

{
  "mcpServers": {
    "rag-server": {
      "command": "uvx",
      "args": [
        "--from",
        "git+https://github.com/yourusername/easy_mcp_rag.git",
        "easy_mcp_rag",
        "--data-dir",
        "/path/to/your/documents",
        "--device",
        "auto",
        "--verbose"
      ]
    }
  }
}

With HTTP Transport

{
  "mcpServers": {
    "rag-server": {
      "command": "uvx",
      "args": [
        "--from",
        "git+https://github.com/yourusername/easy_mcp_rag.git",
        "easy_mcp_rag",
        "--data-dir",
        "/path/to/your/documents",
        "--transport",
        "http",
        "--http-port",
        "8000"
      ]
    }
  }
}

🛠️ How It Works

1. Scan - Discovers all subdirectories in your data directory
2. Load - Extracts text from all supported file types
3. Chunk - Splits documents into overlapping chunks
4. Embed - Generates vector embeddings (CPU or GPU)
5. Index - Stores in Qdrant (one collection per subdirectory)
6. Serve - Creates MCP tools for each collection

Example

documents/
├── legal_docs/      → Creates "legal_docs_search" tool
├── research/        → Creates "research_search" tool
└── data/            → Creates "data_search" tool

📄 Supported File Types

Category	Extensions
Text	.txt, .md, .py, .js, .json, .xml, .html, .css
PDF	.pdf
Word	.docx, .doc
Spreadsheet	.csv, .xlsx, .xls

🎨 Embedding Models

Choose based on your needs:

Model	Dimensions	Speed	Quality	Use Case
all-MiniLM-L6-v2	384	⚡⚡⚡	Good	Default, fast
all-MiniLM-L12-v2	384	⚡⚡	Better	Balanced
all-mpnet-base-v2	768	⚡	Best	Quality

🐛 Troubleshooting

Qdrant Connection Failed

# Check if Qdrant is running
curl http://localhost:6333

# Start Qdrant
docker run -p 6333:6333 qdrant/qdrant

GPU Not Detected

# Check PyTorch GPU support
python -c "import torch; print(torch.cuda.is_available())"

# Install with GPU support
uv pip install -e ".[gpu]"

Out of Memory

# Use smaller model
--embedding-model all-MiniLM-L6-v2

# Reduce batch size
--batch-size 16

# Use CPU
--device cpu

📊 Logging

Enable verbose logging to see detailed information:

easy_mcp_rag --data-dir ./documents --verbose

Output includes:

• ✅ Tool access events
• 🔍 Query details
• 📈 Result counts
• 🎯 Relevance scores
• 📁 Source files

Example:

2024-01-20 10:30:15 - easy_mcp_rag.server - INFO - Tool accessed: legal_docs_search
2024-01-20 10:30:15 - easy_mcp_rag.server - INFO - Query: contract terms
2024-01-20 10:30:15 - easy_mcp_rag.server - INFO - Results returned: 5
2024-01-20 10:30:15 - easy_mcp_rag.server - DEBUG - Result 1: score=0.8542

🔐 Security Notes

• HTTP mode exposes the server on the network
• Use --http-host 127.0.0.1 for local-only access
• Consider authentication for production deployments

📝 Development

# Clone repository
git clone https://github.com/yourusername/easy_mcp_rag.git
cd easy_mcp_rag

# Install with dev dependencies
uv pip install -e ".[dev]"

# Run tests
pytest

# Format code
black src/

# Lint
ruff src/

🤝 Contributing

Contributions welcome! Please:

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Submit a pull request

📜 License

MIT License - see LICENSE file

🙏 Credits

Built with:

• MCP - Model Context Protocol
• Qdrant - Vector database
• Sentence Transformers - Embeddings
• UV - Package manager