MCP Server Poc

🚀 MCP Server POC

A proof-of-concept implementation of a Model Context Protocol (MCP) server for enhancing the capabilities of AI assistants through custom tools and resources.

🚀 Quick Start

This project demonstrates how to create and use a Model Context Protocol (MCP) server. The server can provide custom tools and resources for AI assistants that support the MCP standard, such as Claude. The server includes:

• Document search tools from LangChain, LlamaIndex, and OpenAI.
• Web scraping functionality.
• Integration with the Google Search API.

✨ Features

Tools

• get_docs(query, library): Search the documentation of a specified library (langchain, llama-index, or openai) and return relevant information.

📦 Installation

Prerequisites

• Python 3.11
• Packages listed in requirements.txt

Steps

1. Clone this repository:

git clone https://github.com/yourusername/mcp-server-poc.git
cd mcp-server-poc

2. Create and activate a virtual environment:

# Initialize the uv package manager 
uv init

# Create a Python 3.11 virtual environment
uv venv

# Activate on Windows
.venv\Scripts\activate

# Activate on macOS/Linux
source venv/bin/activate

3. Install the required packages:

uv pip install -r requirements.txt

4. (Optional) If you need development or alternative integration, install the MCP CLI tool:

uv add "mcp[cli]"

5. Set up environment variables: Create a .env file in the root directory with the following content:

SERPER_API_KEY=your_serper_api_key_here

💻 Usage Examples

Basic Usage

To run the MCP server:

uv run mcp 

The server will start and wait for connections using the stdio transport method.

Advanced Usage

Integration with Cursor

To use this MCP server with the Cursor IDE:

1. Create or edit the file ~/.cursor/mcp.json (on Windows: C:\Users\<username>\.cursor\mcp.json) with the following content:

{
    "mcpServers": {
        "mcp-server": {
            "command": "python", 
            "args": [
                "ABSOLUTE/PATH/TO/main.py"
            ]
        }
    }
}

2. Replace the path with the absolute path to the main.py file.
   • On Windows, use double backslashes: C:\\Users\\username\\path\\to\\main.py
   • On macOS/Linux, use regular slashes: /Users/username/path/to/main.py
3. Completely restart Cursor (including ending any background processes) to load the MCP server.

🔧 Technical Details

Windows Binary Mode Fix

This server includes a specific fix to ensure correct operation with stdio transport on Windows:

# Set binary mode for (stdin/stdout) on Windows
if os.name == 'nt':
    import msvcrt
    msvcrt.setmode(sys.stdin.fileno(), os.O_BINARY)
    msvcrt.setmode(sys.stdout.fileno(), os.O_BINARY)

This fix is necessary because Windows differentiates between text and binary mode file handling, which can cause issues with the stdio transport mechanism.

📚 Documentation

Troubleshooting

Windows-specific issues

• “Failed to create client” or “Client closed” errors:
  • Ensure that the included binary mode fix is used.
  • Use absolute paths with double backslashes in the configuration file.
  • Try running the MCP server directly to view any error output.
  • Completely exit Cursor (including terminating any background processes via the Task Manager) and then restart.

General issues

• Verify that all required packages are installed (check using pip list).
• Check that the absolute paths in the configuration file are correct.
• Ensure that the MCP server uses the correct Python version (3.11).
• Verify that the .env file contains the required API keys.
• Try reinstalling the MCP package: pip uninstall mcp && pip install mcp

📄 License

MIT License

Acknowledgments

• This project uses the Model Context Protocol SDK.
• The web scraping functionality is provided by crawl4ai.