medonomator/claude-rag-mcp

Claude RAG MCP Server

MCP server for Claude Code with RAG (Retrieval-Augmented Generation) capabilities. Automatically saves and indexes your coding sessions for semantic search.

Features

• Semantic Search - Find solutions by meaning, not just keywords
• Auto-chunking - Intelligent text splitting with overlap
• Project Isolation - Data organized by project
• Flexible Storage - Qdrant (vectors) + SQLite (metadata)
• Multiple Embedding Providers - OpenAI or Ollama (local)
• Knowledge Types - Solutions, commands, code, notes, bugfixes, architecture

Quick Start

1. Prerequisites

• Node.js 20+
• Docker (for Qdrant)
• OpenAI API key (or Ollama for local embeddings)

2. Start Qdrant

cd docker
docker-compose up -d

3. Install & Build

npm install
npm run build

4. Configure Claude Code

Add to ~/.claude/settings.json:

{
  "mcpServers": {
    "rag": {
      "command": "node",
      "args": ["/path/to/claude-rag-mcp/dist/index.js"],
      "env": {
        "EMBEDDING_PROVIDER": "openai",
        "OPENAI_API_KEY": "sk-your-key",
        "QDRANT_URL": "http://localhost:6333"
      }
    }
  }
}

5. Restart Claude Code

Usage

Save Knowledge

rag_save: Save a solution about XYZ
- content: "To fix the simulator issue, run xcrun simctl boot <UDID>"
- type: "solution"
- project_id: "confyday-ios"
- tags: ["xcode", "simulator"]

Search Knowledge

rag_search: Find solutions about simulators
- query: "how to fix simulator"
- project_id: "confyday-ios"

List Entries

rag_list: Show all saved knowledge
- project_id: "confyday-ios"
- type: "solution"

Delete Entry

rag_delete: Remove entry
- id: "uuid-here"

Tools Reference

Configuration

Environment Variables

# Embedding Provider: openai | ollama
EMBEDDING_PROVIDER=openai

# OpenAI
OPENAI_API_KEY=sk-...
EMBEDDING_MODEL=text-embedding-3-small  # or text-embedding-3-large
EMBEDDING_DIMENSIONS=1536               # 1536 for small, 3072 for large

# Ollama (local, free)
OLLAMA_URL=http://localhost:11434
OLLAMA_MODEL=nomic-embed-text

# Qdrant
QDRANT_URL=http://localhost:6333
QDRANT_COLLECTION=claude_rag

# SQLite
SQLITE_PATH=~/.claude-rag/data.db

# Chunking
CHUNK_SIZE=512
CHUNK_OVERLAP=50

# Search
DEFAULT_SEARCH_LIMIT=10
MIN_SCORE_THRESHOLD=0.5

# Logging
LOG_LEVEL=info  # debug | info | warn | error

Using Ollama (Free, Local)

1. Install Ollama: https://ollama.ai
2. Pull embedding model:

   ollama pull nomic-embed-text
   

3. Set environment:

   EMBEDDING_PROVIDER=ollama
   OLLAMA_URL=http://localhost:11434
   OLLAMA_MODEL=nomic-embed-text
   

Architecture

┌─────────────────────────────────────────┐
│            Claude Code                   │
│  rag_save()  rag_search()  rag_list()   │
└──────────────────┬──────────────────────┘
                   │ stdio (JSON-RPC)
┌──────────────────▼──────────────────────┐
│          MCP RAG Server                  │
│  ┌──────────┐  ┌──────────┐  ┌────────┐ │
│  │ Chunker  │  │ Embedder │  │ Tools  │ │
│  └──────────┘  └──────────┘  └────────┘ │
└──────────────────┬──────────────────────┘
        ┌──────────┴──────────┐
        │                     │
┌───────▼───────┐    ┌───────▼───────┐
│    Qdrant     │    │    SQLite     │
│   (vectors)   │    │  (metadata)   │
└───────────────┘    └───────────────┘

Development

# Watch mode
npm run dev

# Run tests
npm test

# Lint
npm run lint

# Format
npm run format

Cost Estimation

For typical usage (~100K tokens/month): < $0.01/month with OpenAI small.

License

MIT

Links

• MCP Documentation
• Qdrant Documentation
• OpenAI Embeddings