NestJS MCP Server - Model Context Protocol Example

By: @LiusDev

This repository demonstrates a NestJS implementation of the Model Context Protocol (MCP) with a microservice architecture. It consists of two main services:

1. mcp-server: Provides functions to get current time context for LLMs
2. mcp-backend: A client that uses LangChain.js and integrates with the MCP client SDK to connect to the MCP server

Getting Started

Prerequisites

• Node.js (v20 or higher)
• npm

Installation

1. Clone the repository
2. Install dependencies:

npm install

1. Rename the .env.example to .env and add your OpenAI API key:

cp .env.example .env

Then edit the .env file to include your OpenAI API key:

OPENAI_API_KEY=your_openai_api_key_here
OPENAI_API_URL=https://api.openai.com/v1
PORT=3001

Running the Services

You need to run both services for the complete functionality:

Start the MCP Server

npm run start:dev mcp-server

This will start the MCP server on port 3000 (default).

Start the MCP Backend

npm run start:dev mcp-backend

This will start the MCP backend on port 3001 (default).

Usage Example

Once both services are running, you can test the functionality by sending a POST request to the MCP backend:

Sample Request

Send a POST request to http://localhost:3001 with the following JSON body:

{
  "message": "What time is it in Viet Nam?"
}

Using cURL

curl -X POST http://localhost:3001 -H "Content-Type: application/json" -d "{\"message\": \"What time is it in Viet Nam?\"}"

Using Postman

1. Create a new POST request to http://localhost:3001
2. Set the Content-Type header to application/json
3. In the request body, select "raw" and "JSON", then enter:

   {
     "message": "What time is it in Viet Nam?"
   }
   

4. Send the request

The response will contain the current time in Vietnam, retrieved through the MCP server's time context function.

Connecting to Multiple MCP Servers

The backend can connect to multiple MCP servers simultaneously. To add additional servers, modify the McpClientModule.register configuration in apps/mcp-backend/src/mcp-backend.module.ts:

McpClientModule.register({
  throwOnLoadError: true,
  prefixToolNameWithServerName: false,  // Set to true to prefix tool names with server names
  additionalToolNamePrefix: '',
  mcpServers: {
    myServer: {
      transport: 'sse',
      url: 'http://localhost:3000/sse',
      useNodeEventSource: true,
      reconnect: {
        enabled: true,
        maxAttempts: 5,
        delayMs: 2000,
      },
    },
    // Add additional servers here
    anotherServer: {
      transport: 'sse',
      url: 'http://localhost:4000/sse',  // Different port for another server
      useNodeEventSource: true,
      reconnect: {
        enabled: true,
        maxAttempts: 5,
        delayMs: 2000,
      },
    },
  },
}),

When connecting to multiple servers:

• Consider setting prefixToolNameWithServerName: true to avoid tool name conflicts
• Ensure each server has a unique key in the mcpServers object
• Make sure each server is running on a different port

The MCP client will automatically fetch tools from all configured servers and make them available to the LLM.

Architecture

• mcp-server: Exposes tools via the Model Context Protocol, including a function to get the current time
• mcp-backend: Connects to the MCP server, retrieves available tools, and uses them with LangChain.js to process user queries

Technologies Used

• NestJS
• LangChain.js
• Model Context Protocol (MCP)
• @langchain/mcp-adapters: MCP client adapters for LangChain.js
• @rekog/mcp-nest: MCP server implementation for NestJS

Project Structure

mcp-server/
├── apps/
│   ├── mcp-server/       # MCP server implementation
│   └── mcp-backend/      # MCP client implementation
├── dist/                 # Compiled output
├── node_modules/
├── .env                  # Environment variables
└── package.json

License

This project is licensed under the UNLICENSED License - see the LICENSE file for details.