Ghost OS, developed by Ghostwright, is an MCP server tool for macOS 14+ built with Swift 6.2. It enables AI agents to interact with a computer environment, providing full system control. Distributed via Homebrew, direct binary download, or source build, Ghost OS connects to various AI agents through the Model Context Protocol. It allows AI to perform complex actions and automate workflows directly on a macOS system, acting as an AI agent skill.

What does Ghost OS connect to?

Ghost OS connects to AI agents such as Claude Code, Codex CLI, ChatGPT, and Cursor using the Model Context Protocol (MCP). It functions as an MCP server or an AI agent skill, providing AI agents with full computer-use capabilities on macOS. Communication occurs via transport protocols that differentiate between Content-Length, used by clients like Claude Code, and NDJSON (Newline Delimited JSON), used by clients like Claude Desktop. It is compatible with any other MCP client.

Core features and tools

Ghost OS exposes a set of callable tools for AI agent interaction. These include UI interaction commands such as ghost_click, ghost_long_press, ghost_drag, ghost_type, ghost_press, ghost_hotkey, ghost_scroll, ghost_focus, and ghost_window. It also provides workflow management tools like ghost_wait, ghost_recipes, ghost_run, ghost_recipe_show, ghost_recipe_save, ghost_recipe_delete, ghost_learn_start, ghost_learn_stop, and ghost_learn_status for learning and executing tasks.

Who is it for?

Ghost OS is for AI agents and developers aiming to equip AI agents with extensive computer-use capabilities on macOS. It allows AI to perform actions like downloading papers, sending emails, and managing UI elements across applications. This tool is designed for those who need to grant AI direct control over a macOS environment for task automation and interaction. Skip this if your primary need is not to grant AI agents direct macOS control.

How to get started

No credentials are required. Setup involves running "ghost setup" which handles MCP configuration, recipe installation, vision model setup, and prompts for required macOS permissions.

Use cases

• Downloading a paper from arXiv using a recipe, e.g., "ghost_run recipe:"arxiv-download" params:{query:"chain of thought prompting"}".
• Sending an email by learning a workflow from user actions, then replaying it with parameters like recipient and subject.
• General UI interaction including clicking buttons, typing text, dragging elements, scrolling, and managing windows on macOS.

Ghost OS empowers AI agents with direct macOS interaction, enabling them to perform a variety of computer-use tasks. Its toolset supports both direct UI manipulation and complex workflow automation through recipes. The system connects with multiple AI clients using the Model Context Protocol, making it a valuable asset for AI-driven automation on macOS.