macOS Use

An intelligent agent for macOS that uses LLMs to control your computer. Built on top of macos-mcp, it provides a powerful interface for automating tasks on your Mac.

Features

• Natural Language Control: control your Mac using plain English commands.
• Accessibility Tree Integration: Uses advanced accessibility tree traversal (based on macos-mcp) to understand UI structure.
• Smart Window Management: Can launch, switch, resize, and manage application windows.
• Robust Tooling: Includes tools for clicking, typing, scrolling, and more, with built-in reliability features.
• Agentic Workflow: Uses an autonomous agent loop to plan and execute complex multi-step tasks.

Installation

1. Clone the repository:

   git clone https://github.com/Jeomon/macos-use.git
   cd macos-use

2. Install dependencies using uv or pip:

   pip install -e .

3. Configure environment variables: Create a .env file with your API keys (e.g., ANTHROPIC_API_KEY).

Usage

Run the agent:

python main.py

Then enter your command, for example:

• "Open Safari and search for 'latest AI news'"
• "Organize my windows side by side"
• "Check my calendar for upcoming meetings"

Architecture

This project integrates:

• Agent Core: Handles reasoning, planning, and tool execution.
• Desktop Service: Manages screen capture, windowing, and input simulation.
• Tree Service: Efficiently traverses the macOS accessibility tree to find interactive elements (imported from macos-mcp).

License

MIT