This project implements a Model Context Protocol (MCP) server in Go that provides filesystem access to files in /tmp/mcp, and a client that uses OpenAI's GPT-4o-mini model to interact with the filesystem tools.

• MCP Server: Implements the Model Context Protocol to provide filesystem tools
• Filesystem Tools:
  • read_file: Read contents of files in /tmp/mcp
  • list_files: List files and directories in /tmp/mcp
• AI Client: Uses OpenAI GPT-4o-mini to intelligently interact with filesystem tools
• Security: Restricted to /tmp/mcp directory for safety

• Go 1.21 or later
• OpenAI API key

1. Clone the repository:

git clone <repository-url>
cd mcp-test

2. Install dependencies:

make deps

3. Build the server and client:

make build

Set your OpenAI API key as an environment variable:

export OPENAI_API_KEY="your-openai-api-key-here"

The client will automatically start the MCP server and provide an interactive chat interface:

make run

Or run the client directly:

./client

Once the client is running, you can interact with it using natural language:

> List all files in the directory
> Read the contents of file.txt
> What files are in the subdirectory?
> Show me the contents of config.json

You can also test the server directly by creating some test files:

# Create the test directory
mkdir -p /tmp/mcp

# Create some test files
echo "Hello, World!" > /tmp/mcp/test.txt
echo '{"name": "test", "value": 42}' > /tmp/mcp/config.json
mkdir /tmp/mcp/subdir
echo "Subdirectory content" > /tmp/mcp/subdir/file.txt

The server implements the Model Context Protocol and provides two tools:

1. read_file: Reads file contents from /tmp/mcp

   • Parameters: path (string) - relative path to file
   • Returns: File contents as text

2. list_files: Lists directory contents in /tmp/mcp

   • Parameters: path (string, optional) - relative path to directory
   • Returns: List of files and directories

The client:

• Starts the MCP server as a subprocess
• Communicates with the server using JSON-RPC over stdin/stdout
• Uses OpenAI GPT-4o-mini to interpret user requests and call appropriate tools
• Provides an interactive chat interface

• The server is restricted to only access files within /tmp/mcp
• Path traversal attacks are prevented by checking that all paths start with /tmp/mcp
• The server runs as a subprocess and can be easily terminated

# Build only the server
make server

# Build only the client
make client

make clean

mcp-test/
├── server/
│   └── main.go          # MCP server implementation
├── client/
│   └── main.go          # MCP client with OpenAI integration
├── go.mod               # Go module dependencies
├── Makefile             # Build and run commands
└── README.md            # This file

1. "OPENAI_API_KEY environment variable is required"

   • Make sure you've set the environment variable before running the client

2. "Failed to start server"

   • Ensure the server binary was built successfully with make build

3. "File not found" errors

   • Make sure the files exist in /tmp/mcp directory
   • Check that the path is relative to /tmp/mcp

4. Permission errors

   • Ensure you have read permissions for files in /tmp/mcp
   • The client will create /tmp/mcp if it doesn't exist

To see more detailed logging, you can modify the log level in the source code or add debug prints.

This project is provided as-is for educational and development purposes.