@laskarks/mcp Rag Node

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"],
      "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"],
      "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

Installation

npm i @laskarks/mcp-rag-node

Environment Variables

Required

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.

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

Optional

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