mcpField
C
click-mcp

click-mcp

Local
github.com/crowecawcawDeveloper Tools14Updated 14w ago

About click-mcp

click-mcp is a Python library that turns commands in Click applications into MCP tools for AI agents to invoke CLI operations. It supports nested command groups and positional arguments, and serves MCP over stdio.

Key features

  1. 01Convert Click commands with a decorator
  2. 02Support nested command groups
  3. 03Support positional arguments
  4. 04Run an MCP server over stdio
  5. 05Return command errors in MCP responses

Use cases

  1. 01Let AI agents invoke existing Click CLIs
  2. 02Expose nested subcommands as MCP tools

MCP support

The library adds an MCP server command to Click applications and maps their commands to MCP tools for clients to discover and invoke. The server communicates over stdio.

Configuration

Local server config is the same for every client. Copy it into your client’s MCP config file.

No install method listed yet

We haven’t collected a package or remote endpoint for this server. Follow the instructions in the project README.

Repository

Tools

No tool list yet

Local servers expose tools only once running on your machine, and the maintainer has not provided a tool list.

Details

From the project README

click-mcp

PyPI version License: MIT

A Python library that extends Click applications with Model Context Protocol (MCP) support, allowing AI agents to interact with CLI tools.

Overview

click-mcp provides a simple decorator that converts Click commands into MCP tools. This enables AI agents to discover and interact with your CLI applications programmatically.

The Model Context Protocol (MCP) is an open standard for AI agents to interact with tools and applications in a structured way.

Key Features

  • Simple @click_mcp decorator syntax
  • Automatic conversion of Click commands to MCP tools
  • Support for nested command groups
  • Support for positional arguments
  • Stdio-based MCP server for easy integration

Installation

pip install click-mcp

Basic Usage

import click
from click_mcp import click_mcp

@click_mcp(server_name="my-cli-app")
@click.group()
def cli():
    """Sample CLI application."""
    pass

@cli.command()
@click.option('--name', required=True, help='Name to greet')
def greet(name):
    """Greet someone."""
    click.echo(f"Hello, {name}!")

if __name__ == '__main__':
    cli()

When you run the MCP server, Click commands are converted into MCP tools:

  • Command greet becomes MCP tool greet
  • Nested commands use dot notation (e.g., users.create)

To invoke a command via MCP, send a request like:

{
  "type": "invoke",
  "tool": "greet",
  "parameters": {
    "name": "World"
  }
}

To start the MCP server:

$ python my_app.py mcp

Advanced Usage

Customizing the MCP Command Name

By default, click-mcp adds an mcp command to your CLI application. You can customize this name using the command_name parameter:

@click_mcp(command_name="start-mcp")
@click.group()
def cli():
    """Sample CLI application with custom MCP command name."""
    pass

With this configuration, you would start the MCP server using:

$ python my_app.py start-mcp

This can be useful when:

  • The name "mcp" conflicts with an existing command
  • You want a more descriptive command name
  • You're integrating with a specific AI agent that expects a certain command name
Customizing the MCP Server Name

You can also customize the name of the MCP server that's reported to clients:

@click_mcp(server_name="my-custom-tool")
@click.group()
def cli():
    """Sample CLI application with custom server name."""
    pass

This can be useful when:

  • You want to provide a more descriptive name for your tool
  • You're integrating with systems that use the server name for identification
  • You want to distinguish between different MCP-enabled applications
Working with Nested Command Groups

click-mcp supports nested command groups. When you have a complex CLI structure with subcommands, all commands are exposed as MCP tools:

@click_mcp
@click.group()
def cli():
    """Main CLI application."""
    pass

@cli.group()
def users():
    """User management commands."""
    pass

@users.command()
@click.option('--username', required=True)
def create(username):
    """Create a new user."""
    click.echo(f"Creating user: {username}")

@users.command()
@click.argument('username')
def delete(username):
    """Delete a user."""
    click.echo(f"Deleting user: {username}")

When exposed as MCP tools, the nested commands will be available with their full path using dot notation (e.g., "users.create" and "users.delete").

Working with Positional Arguments

Click supports positional arguments using @click.argument(). When these are converted to MCP tools, they are represented as named parameters in the schema:

@cli.command()
@click.argument('source')
@click.argument('destination')
@click.option('--overwrite', is_flag=True, help='Overwrite destination if it exists')
def copy(source, destination, overwrite):
    """Copy a file from source to destination."""
    click.echo(f"Copying {source} to {destination}")

This command is converted to an MCP tool with the following schema:

{
  "type": "object",
  "properties": {
    "source": {
      "description": "",
      "schema": { "type": "string" },
      "required": true
    },
    "destination": {
      "description": "",
      "schema": { "type": "string" },
      "required": true
    },
    "overwrite": {
      "description": "Overwrite destination if it exists",
      "schema": { "type": "boolean" }
    }
  },
  "required": ["source", "destination"]
}

The positional nature of arguments is handled internally by click-mcp. When invoking the command, you can use named parameters:

{
  "type": "invoke",
  "tool": "copy",
  "parameters": {
    "source": "file.txt",
    "destination": "/tmp/file.txt",
    "overwrite": true
  }
}

The MCP server will correctly convert these to positional arguments when executing the Click command:

copy file.txt /tmp/file.txt --overwrite
Handling Command Errors

When a Click command raises an exception, click-mcp captures the error and returns it as part of the MCP response. This allows AI agents to handle errors gracefully:

@cli.command()
@click.option('--filename', required=True)
def process(filename):
    """Process a file."""
    try:
        with open(filename, 'r') as f:
            content = f.read()
        click.echo(f"Processed file: {filename}")
    except FileNotFoundError:
        raise click.UsageError(f"File not found: {filename}")

If the file doesn't exist, the AI agent will receive an error message that it can present to the user or use to take corrective action.

Development

Setup

Clone the repository and install Hatch:

git clone https://github.com/aws/click-mcp.git
cd click-mcp
pip install hatch
Testing

Run tests with Hatch:

# Run all tests
hatch run test

# Run tests with coverage
hatch run cov
Code Formatting

Format code with Black:

# Format code
hatch run format

# Check formatting
hatch run check-format
Linting

Run linting checks with Ruff:

hatch run lint
Type Checking

Run type checking with MyPy:

hatch run typecheck
Run All Checks

Run all checks (formatting, linting, type checking, and tests):

hatch run check-all
Building

Build the package:

hatch run build
Documentation

Generate documentation:

hatch run docs

Related Resources

License

MIT

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.