golf-mcp 0.1.4

Framework for building MCP servers

⛳ Golf

Easiest framework for building MCP servers

📚 Documentation

Overview

Golf is a framework designed to streamline the creation of MCP server applications. It allows developers to define server's capabilities—tools, prompts, and resources—as simple Python files within a conventional directory structure. Golf then automatically discovers, parses, and compiles these components into a runnable FastMCP server, minimizing boilerplate and accelerating development.

With Golf, you can focus on implementing your agent's logic rather than wrestling with server setup and integration complexities. It's built for developers who want a quick, organized way to build powerful MCP servers.

Quick Start

Get your Golf project up and running in a few simple steps:

1. Install Golf

Ensure you have Python (3.10+ recommended) installed. Then, install Golf using pip:

pip install golf-mcp

2. Initialize Your Project

Use the Golf CLI to scaffold a new project:

golf init your-project-name

This command creates a new directory (your-project-name) with a basic project structure, including example tools, resources, and a golf.json configuration file.

3. Run the Development Server

Navigate into your new project directory and start the development server:

cd your-project-name
golf build dev
golf run

This will start the FastMCP server, typically on http://127.0.0.1:3000 (configurable in golf.json).

That's it! Your Golf server is running and ready for integration.

Basic Project Structure

A Golf project initialized with golf init will have a structure similar to this:

<your-project-name>/
│
├─ golf.json          # Main project configuration
│
├─ tools/             # Directory for tool implementations
│   └─ hello.py       # Example tool
│
├─ resources/         # Directory for resource implementations
│   └─ info.py        # Example resource
│
├─ prompts/           # Directory for prompt templates
│   └─ welcome.py     # Example prompt
│
├─ .env               # Environment variables (e.g., API keys, server port)
└─ pre_build.py       # (Optional) Script for pre-build hooks (e.g., auth setup)

• golf.json: Configures server name, port, transport, telemetry, and other build settings.
• tools/, resources/, prompts/: Contain your Python files, each defining a single component. These directories can also contain nested subdirectories to further organize your components (e.g., tools/payments/charge.py). The module docstring of each file serves as the component's description.
  • Component IDs are automatically derived from their file path. For example, tools/hello.py becomes hello, and a nested file like tools/payments/submit.py would become submit-payments (filename, followed by reversed parent directories under the main category, joined by hyphens).
• common.py (not shown, but can be placed in subdirectories like tools/payments/common.py): Used to share code (clients, models, etc.) among components in the same subdirectory.

Example: Defining a Tool

Creating a new tool is as simple as adding a Python file to the tools/ directory. The example tools/hello.py in the boilerplate looks like this:

# tools/hello.py
"""Hello World tool {{project_name}}."""

from pydantic import BaseModel

class Output(BaseModel):
    """Response from the hello tool."""
    message: str

async def hello(
    name: str = "World",
    greeting: str = "Hello"
) -> Output:
    """Say hello to the given name.
    
    This is a simple example tool that demonstrates the basic structure
    of a tool implementation in Golf.
    """
    print(f"{greeting} {name}...")
    return Output(message=f"{greeting}, {name}!")

# Designate the entry point function
export = hello

Golf will automatically discover this file. The module docstring """Hello World tool {{project_name}}.""" is used as the tool's description. It infers parameters from the hello function's signature and uses the Output Pydantic model for the output schema. The tool will be registered with the ID hello.

Configuration (golf.json)

The golf.json file is the heart of your Golf project configuration. Here's what each field controls:

{
  "name": "{{project_name}}",          // Your MCP server name (required)
  "description": "A Golf project",     // Brief description of your server's purpose
  "host": "127.0.0.1",                // Server host - use "0.0.0.0" to listen on all interfaces
  "port": 3000,                       // Server port - any available port number
  "transport": "sse",                 // Communication protocol:
                                      // - "sse": Server-Sent Events (recommended for web clients)
                                      // - "streamable-http": HTTP with streaming support
                                      // - "stdio": Standard I/O (for CLI integration)
}

Key Configuration Options:

• name: The identifier for your MCP server. This will be shown to clients connecting to your server.
• transport: Choose based on your client needs:
  • "sse" is ideal for web-based clients and real-time communication
  • "streamable-http" provides HTTP streaming for traditional API clients
  • "stdio" enables integration with command-line tools and scripts
• host & port: Control where your server listens. Use "127.0.0.1" for local development or "0.0.0.0" to accept external connections.

Privacy & Telemetry

Golf collects anonymous usage data to help us understand how the framework is being used and improve it over time. The data collected includes:

• Commands run (init, build, run)
• Success/failure status (no error details)
• Golf version, Python version (major.minor only), and OS type
• Template name (for init command only)
• Build environment (dev/prod for build commands only)

No personal information, project names, code content, or error messages are ever collected.

Opting Out

You can disable telemetry in several ways:

1. Using the telemetry command (recommended):

   golf telemetry disable
   

   This saves your preference permanently. To re-enable:

   golf telemetry enable
   

2. During any command: Add --no-telemetry to save your preference:

   golf init my-project --no-telemetry
   

3. Environment variable (temporary override):

   export GOLF_TELEMETRY=0
   golf init my-project
   

Your telemetry preference is stored in ~/.golf/telemetry.json and persists across all Golf commands.

Roadmap

Here are the things we are working hard on:

• Native OpenTelemetry implementation for tracing
• golf deploy command for one click deployments to Vercel, Blaxel and other providers
• Production-ready OAuth token management, to allow for persistent, encrypted token storage and client mapping

Made with ❤️ in Warsaw, Poland and SF