phone-harness is a Python CLI that connects an LLM agent to a real iPhone via the macOS iPhone Mirroring window, using Vision-framework OCR for screen reading and HID-level CGEvents for taps, drags, and keystrokes.

It requires macOS Sequoia, one-time iPhone pairing, and Accessibility/Screen Recording permissions; no jailbreak, Xcode, or WebDriverAgent is needed. The stateless tool exposes primitives like open_app, tap_text, and type_text, with limits including no multi-touch, no camera or Face ID flows, and OCR-only text recognition without semantic parsing.