# In Claude Code, after configuring the server:
"Search my knowledge base for articles about RAG architecture"
# Claude automatically uses rag_query tool and gets relevant context

┌──────────────┐     MCP Protocol     ┌──────────────┐
│  MCP Client  │◀────────────────────▶│  MCP Server  │
│ (Claude Code)│                      │ (This repo)  │
└──────────────┘                      └──────────────┘
                                             │
                                      ┌──────▼──────┐
                                      │ RAG Pipeline │
                                      │  (ChromaDB)  │
                                      └─────────────┘

# Start ChromaDB
docker run -p 8000:8000 chromadb/chroma

# Set OpenAI API key (for embeddings)
export OPENAI_API_KEY="sk-..."

git clone https://github.com/0xrdan/mcp-rag-server.git
cd mcp-rag-server
npm install
npm run build

{
  "mcpServers": {
    "rag": {
      "command": "node",
      "args": ["/path/to/mcp-rag-server/dist/server.js"],
      "env": {
        "OPENAI_API_KEY": "sk-...",
        "CHROMA_URL": "http://localhost:8000",
        "CHROMA_COLLECTION": "my_knowledge_base"
      }
    }
  }
}

# Restart Claude Code to load the server
claude

# Now Claude has access to RAG tools:
"Index this document into my knowledge base: [paste content]"
"Search for information about transformer architectures"
"What do my docs say about error handling?"

// Input
{
  question: string;      // Required: the question to search for
  topK?: number;         // Optional: number of results (default: 5)
  threshold?: number;    // Optional: min similarity 0-1 (default: 0.5)
  filters?: object;      // Optional: metadata filters
}

// Output
{
  context: string;       // Formatted context for LLM
  chunks: [{
    content: string;
    score: number;
    metadata: object;
  }];
  stats: {
    totalChunks: number;
    avgSimilarity: number;
  };
}

// Input
{
  query: string;         // Required: search query
  topK?: number;         // Optional: number of results (default: 10)
  filters?: object;      // Optional: metadata filters
}

// Output: Array of chunks with scores

// Input
{
  id: string;            // Required: unique identifier
  title: string;         // Required: document title
  content: string;       // Required: document content
  source: string;        // Required: source identifier
  category?: string;     // Optional: category
  tags?: string[];       // Optional: tags array
}

// Output
{
  success: boolean;
  documentId: string;
  chunksIndexed: number;
}

// Output
{
  totalChunks: number;
  totalDocuments: number;
  // ... other stats from RAG pipeline
}

mcp-rag-server/
├── src/
│   ├── server.ts        # Main MCP server implementation
│   └── index.ts         # Exports
├── mcp-config.example.json  # Example Claude Code configuration
├── package.json
└── README.md

import { createServer } from 'mcp-rag-server';

const server = await createServer({
  vectorDB: {
    host: 'http://custom-chroma:8000',
    collectionName: 'my_collection',
  },
  rag: {
    topK: 10,
    enableHybridSearch: true,
  },
});

// ~/Library/Application Support/Claude/claude_desktop_config.json (macOS)
{
  "mcpServers": {
    "rag": {
      "command": "node",
      "args": ["/path/to/mcp-rag-server/dist/server.js"]
    }
  }
}

Co-Authored-By: Claude <noreply@anthropic.com>