MCP Starter

🚀 MCP Starter Project

The Model Context Protocol (MCP) is a standard for building AI applications that can interact with external tools and APIs.

🚀 Quick Start

Prerequisites

• Python 3.11 or higher
• Node.js 18 or higher
• Hugging Face API key
• Serper API key for Google search functionality

Setting up the Server

1. Create and activate a Python virtual environment:

cd mcp-server
python -m venv .venv
# On Windows
.venv\Scripts\activate

2. Install dependencies:

pip install -e .

3. Create a .env file in the mcp-server directory:

SERPER_API_KEY=your_serper_api_key_here

Setting up the Client

1. Install Node.js dependencies:

cd mcp-clients
npm install

2. Create a .env file in the mcp-clients directory:

HUGGINGFACE_API_KEY=your_huggingface_api_key_here

3. Build the TypeScript code:

npm run build

Running the Application

1. Start the MCP server:

cd mcp-server
python main.py

2. Start the client server in a new terminal:

cd mcp-clients
node build/index.js ../mcp-server/main.py

✨ Features

• Document Search Tool: Search documentation of popular AI libraries:
  • LangChain
  • LlamaIndex
  • OpenAI
• Conversation Management: Maintain chat history per session
• Tool Integration: Seamlessly combine AI model responses with tool calls
• Error Handling: Robust error handling for API calls and tool execution

💻 Usage Examples

Basic Usage

The client exposes two endpoints:

• Health Check: GET http://localhost:3000/health
• Chat: POST http://localhost:3000/chat

Example chat request:

{
  "query": "Search for RAG in the langchain documentation",
  "sessionId": "User123"
}

📚 Documentation

What is MCP?

The Model Context Protocol (MCP) is a standard for building AI applications that can interact with external tools and APIs. It consists of two main components:

1. MCP Server: A Python service that defines and exposes tools/functions that can be called by AI models.
2. MCP Client: A TypeScript/JavaScript client that connects to the MCP server and manages the interaction between AI models and tools.

Project Structure

mcp_starter/
├── mcp-server/           # Python MCP server implementation
│   ├── main.py          # Server with document search tool (with documentation)
│   └── pyproject.toml   # Python dependencies
└── mcp-clients/         # TypeScript MCP client implementation
    ├── index.ts         # Express server with HuggingFace integration
    └── package.json     # Node.js dependencies

How it Works

1. The MCP server defines tools that can be called by AI models.
2. The client connects to the MCP server and retrieves available tools.
3. When a user sends a query:
   • The client formats the conversation history.
   • Sends it to the Hugging Face model.
   • Extracts and executes tool calls from the model response.
   • Returns the final response, including tool results.

Environment Variables

Server

• SERPER_API_KEY: API key for Google search functionality.

Client

• HUGGINGFACE_API_KEY: API key to access Hugging Face models.

📄 License

This project is under the MIT License.