Voices
phone-harness Lets Claude Code Drive a Real iPhone With No Jailbreak, No Xcode and No WebDriverAgent — Just Screen Mirroring, Vision OCR and CGEvents
ShawnPana released phone-harness this week (3,749 likes, 453,889 views on the launch post), an open-source harness that connects an LLM directly to a physical iPhone through the Mac's iPhone Mirroring window. It uses screencapture plus Apple's Vision framework OCR to build a 'poor man's DOM' of every visible string with tap-ready coordinates, then drives the device with HID-level CGEvents for taps, long-presses, drags, flicks, scrolls, unicode typing and app shortcuts. The design point matters more than the demo: it needs no API, no jailbreak and no accessibility automation stack, which means any iOS app becomes agent-addressable today.
Source
↳ Follow the thread