Custom Tools and Resources for AI Assistants via MCP Server

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

Overview

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

• Documentation search tool for LangChain, LlamaIndex, and OpenAI
• Web crawling capability
• Integration with Google Search API

Requirements

• Python 3.11
• Required packages listed in requirements.txt

Installation and Setup

1. Clone this repository:

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

1. Create and activate a virtual environment:

```shell
# Create a Python 3.11 virtual environment
python -m venv venv

# Activate on Windows
.\venv\Scripts\Activate.ps1

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

1. Install the required packages:

shell pip install -r requirements.txt

1. Set up environment variables:

Create a .env file in the root directory with the following:

plaintext SERPER_API_KEY=your_serper_api_key_here

Running the Application

To run the MCP server:

python main.py

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

Integrating with Cursor

To use this MCP server with Cursor IDE:

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

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

1. Replace the path with the absolute path to your main.py file.

2. On Windows, use double backslashes: C:\\Users\\username\\path\\to\\main.py

3. On macOS/Linux, use regular slashes: /Users/username/path/to/main.py

4. Restart Cursor completely (including ending any background processes) to load the MCP server.

Features

Tools

• get_docs(query, library): Searches the documentation for the specified library (langchain, llama-index, or openai) and returns relevant information

Technical Details

Windows Binary Mode Fix

This server includes a specific fix for Windows to ensure proper operation with stdio transport:

# 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 distinguishes between text and binary modes for file handling, which can cause issues with the stdio transport mechanism used by MCP.

Troubleshooting

If you encounter issues with the MCP server:

Windows-Specific Issues

• "Failed to create client" or "Client closed" errors:
• Make sure to use the binary mode fix included in the server
• Use the absolute path with double backslashes in the mcp.json configuration
• Try running the MCP server directly to see if it produces any error output
• Completely exit Cursor (including terminating any background processes via Task Manager) before restarting

General Issues

• Verify that all required packages are installed (pip list to check)
• Check that the absolute path in the configuration file is correct
• Make sure the MCP server is running with the proper version of Python (3.11)
• Verify that your .env file contains the required API key
• Try reinstalling the MCP package: pip uninstall mcp && pip install mcp

License

MIT License

Acknowledgements

• This project uses the Model Context Protocol SDK
• Web crawling functionality provided by crawl4ai