MCP Servers Monorepo

A collection of Model Context Protocol (MCP) servers managed with pnpm and Turborepo.

Structure

mcp/
├── packages/
│   └── openai-mcp/     # OpenAI API MCP Server
├── package.json
├── pnpm-workspace.yaml
└── turbo.json

Getting Started

Prerequisites

• Node.js >= 18
• pnpm >= 9.0
• Docker & Docker Compose (for running servers)

Installation

pnpm install
pnpm build

🚀 Quick Start with CLI

Using the MCP CLI

The project includes a CLI tool for easy server management:

# Start OpenAI MCP server (port 3500)
./mcp -m openai

# Start Gemini MCP server (port 3501)
./mcp -m gemini

# Start all servers
./mcp -m all

# Stop a server
./mcp -m openai --down

# Restart a server
./mcp -m openai --restart

# View server logs
./mcp -m openai --logs

# Show help
./mcp --help

Or using pnpm:

pnpm mcp -m openai

Server Endpoints

Server	Port	Health Check	SSE Endpoint
OpenAI	3500	http://localhost:3500/health	http://localhost:3500/sse
Gemini	3501	http://localhost:3501/health	http://localhost:3501/sse

Configuration

Each server requires its API key. Create a .env file in each package directory:

packages/openai-mcp/.env

OPENAI_API_KEY=sk-your-api-key-here

packages/gemini-mcp/.env

GEMINI_API_KEY=your-api-key-here

Development Mode

# Watch mode for local development
pnpm dev

# Build all packages
pnpm build

Packages

Package	Description
@mcp/openai	OpenAI API integration (GPT, DALL-E, Embeddings, TTS)
@mcp/gemini	Google Gemini API integration (Text, Chat, Vision, Embeddings)

🏗️ Architecture

Dual Mode Support

All MCP servers support two modes:

1. stdio mode (default): For Claude Desktop integration
2. HTTP/SSE mode: For external access via REST API

Set MCP_MODE=http environment variable to enable HTTP mode.

Adding New MCP Servers

1. Create a new directory under packages/
2. Add the required package.json and tsconfig.json
3. Implement your MCP server using @modelcontextprotocol/sdk
4. Add HTTP/SSE support with Express
5. Create Dockerfile and docker-compose.yml
6. Update the mcp CLI script to include the new server

📚 API Documentation

Health Check

curl http://localhost:3500/health

Response:

{
  "status": "ok",
  "server": "openai-mcp"
}

Using MCP Tools

Connect to the SSE endpoint using an MCP client or test with:

curl -X POST http://localhost:3500/sse \
  -H "Content-Type: application/json" \
  -d '{"method": "tools/list"}'

License

MIT