GitHub radar
macOS Harness: Give Your LLM Full Mac Control
Announcing macOS Harness — a small Python wrapper that enables an LLM to interact with any application on Mac by taking screenshots, typing, using mouse and scripting.
Here’s a new project from browser-use — a tiny Python process that exposes 6 low level primitives for interacting with macOS to a LLM: take screenshots of windows, send keyboard/mouse actions to a specific application PID (via Apple Accessibility), access AppleScript, a full browser via CDP, and the local filesystem. During interaction, the model learns and writes the necessary code along the way. No prior tooling needed for any specific application. Only available on macOS, MIT license, anonymous data collection.
Why a vibe-coder should care
Once you set your agent on a Mac task, it will discover what to do on its own, even if there’s no tool for a specific application. All you have to do is to copy & paste the following line into Claude Code and let the agent take care of the rest:
How to install
Copy this and send it to your agent — Claude Code, Codex, any of them:
Install macOS Harness from https://github.com/browser-use/macos-harness with uv on Python 3.12, register the skill printed by `macos-harness skill`, run `macos-harness doctor`, explain any missing macOS permissions and ask me before requesting them.
Mac only — a regular one, no GPU needed; install it as a Python package.
Open on GitHub▌ More finds