RAG MCP Server

MCP (Model Context Protocol) server for RAG (Retrieval-Augmented Generation) using Pinecone, OpenAI-compatible embedding APIs, and the official MCP SDK. Save documents and search by semantic similarity via MCP tools.

Add to MCP clients

Claude Desktop (claude_desktop_config.json):

{
  "mcpServers": {
    "rag": {
      "command": "npx",
      "args": ["-y", "@laskarks/mcp-rag-node@latest"],
      "env": {
        "APIKEY": "sk-...",
        "EMBEDDING_MODEL": "text-embedding-3-small",
        "RAG_CHUNK_MAX_TOKENS": 1536,
        "PINECONE_API_KEY": "...",
        "PINECONE_INDEX": "rag-index",
        "PROVIDER": "openai | openrouter"
      }
    }
  }
}

Cursor (.cursor/mcp.json or MCP settings):

{
  "mcpServers": {
    "rag": {
      "command": "npx",
      "args": ["-y", "@laskarks/mcp-rag-node@latest"],
      "env": {
        "APIKEY": "sk-...",
        "EMBEDDING_MODEL": "text-embedding-3-small",
        "RAG_CHUNK_MAX_TOKENS": 1536,
        "PINECONE_API_KEY": "...",
        "PINECONE_INDEX": "rag-index",
        "PROVIDER": "openai | openrouter"
      }
    }
  }
}

Tools

| Tool | Description | | ------------------------ | ------------------------------------------------------ | | save_to_rag | Chunk text, create embeddings, and save to Pinecone. | | search_document_on_rag | Search documents by keyword using semantic similarity. |

Installation

npm i @laskarks/mcp-rag-node

Environment Variables

Required

| Variable | Description | Example | | ------------------ | ---------------------------------------------------------- | --------------------------------------------------------- | | APIKEY | OpenAI or OpenRouter API key for embeddings | sk-... | | EMBEDDING_MODEL | Embedding model ID | text-embedding-3-small, openai/text-embedding-3-small | | PINECONE_API_KEY | Pinecone API key | ... | | PINECONE_INDEX | Pinecone index name (dimension must match embedding model) | rag-index | | PROVIDER | AI provider (allowed values: openai, openrouter) | openai or openrouter |

Important: Create your Pinecone index with the same dimension as your embedding model.

Embedding models and vector dimensions

Use the Dimension column when creating your Pinecone index.

| Model | Dimension | Provider | | ------------------------------------------------ | --------- | ------------------------ | | text-embedding-3-small | 1536 | OpenAI, OpenRouter | | text-embedding-3-large | 3072 | OpenAI, OpenRouter | | text-embedding-ada-002 | 1536 | OpenAI, OpenRouter | | text-embedding-3-small (with dimensions param) | 512–1536 | OpenAI | | voyage-3 | 1024 | Voyage (via OpenRouter) | | nomic-embed-text-v1.5 | 768 | Nomic (via OpenRouter) | | mistral-embed | 1024 | Mistral (via OpenRouter) | | cohere/embed-english-v3.0 | 1024 | Cohere (via OpenRouter) |

For OpenRouter, use the model ID format, e.g. openai/text-embedding-3-small or voyage/voyage-3.

Optional

| Variable | Description | Default | | ---------------------- | ------------------------------------- | ------- | | RAG_CHUNK_MAX_TOKENS | Max tokens per chunk before embedding | 1536 | | RAG_CHUNK_OVERLAP | Overlap tokens between chunks | 50 |

Usage

Run the server

npm run build
npm start

Or with env file:

# .env
APIKEY=sk-...
EMBEDDING_MODEL=text-embedding-3-small
PINECONE_API_KEY=...
PINECONE_INDEX=rag-index
PROVIDER=openai

npm start

Development

# Install dependencies
npm install

# Build
npm run build

# Run server (from compiled JS)
npm start

# Run server (dev, from TypeScript)
npm run dev

# Run sample client
npm run client

Project structure

src/
├── index.ts    # MCP server entry, tools registration
├── ai.ts       # AI controller (chunking, embeddings, Pinecone)
├── env.ts      # Environment loading
└── client.ts   # Example MCP client for testing
dist/           # Compiled output (after npm run build)

Publish to npm

Before publishing:

1. Add files to package.json to include only dist/ and docs:

 "files": ["dist", "README.md"]

2. Ensure npm run build succeeds and dist/ is committed or built on publish.
3. Add bin entry for npx rag-mcp-nodejs (optional):

 "bin": { "rag-mcp-nodejs": "dist/index.js" }

Note: MCP servers are usually run via node dist/index.js; a bin is optional. 4. Set a unique package name (npm may require scoped name, e.g. @yourname/rag-mcp-nodejs). 5. Add repository, homepage, and engines.node in package.json (optional but recommended).

Requirements

• Node.js >= 18
• Pinecone account
• OpenAI or OpenRouter API key

License

ISC