Phone Harness

Overview

Phone Harness is an open-source tool that lets Claude Code, Codex, or any agent control a real iPhone or Android phone. Agents can tap, type, scroll, and read the screen through the phone interface, without jailbreak, Xcode, or installing an app on the phone.

Key Features

• Agents interact with real phone apps by tapping, typing, scrolling, and reading the screen.
• iPhone control works through the Mac's iPhone Mirroring window, including in the background; Android control works over adb.
• Uses Vision OCR or the phone's accessibility tree to see the screen, with real input events for actions.
• Phone Harness Cloud offers hosted Android and iPhones, real phone numbers with eSIM, persistent logins, and multiple devices.
• Integrations include CLI, MCP server, Hermes Agent, and OpenClaw, with shared helpers across phones.
• Cloud Android can start a fresh phone per session or a saved phone that keeps apps and logins; iPhone control can run over HTTPS.

Use Cases

• Test and use real mobile apps with an AI agent.
• Message and call people through the phone interface.
• Grab verification codes for 2FA.
• Run agent-driven tasks such as booking a ride or interacting with apps on hosted Android or iPhone devices.

Getting Started

• Send the setup prompt to a coding agent such as Claude Code or Codex to install or upgrade the phone-harness CLI with uv on Python 3.12 and register the skill.
• Run phone-harness doctor to check the phone setup.
• For cloud phones, use phone-harness cloud login, phone-harness cloud ls, phone-harness cloud start, and phone-harness cloud stop.

Deployment & Requirements

• iPhone: macOS 15 or later, iPhone Mirroring paired to an iOS 18 or later iPhone, and Accessibility and Screen Recording permissions.
• Android: adb and USB debugging on macOS, Linux, or Windows.
• Cloud Android or iPhone: a Phone Harness Cloud account; cloud Android also needs adb.

Before You Adopt

• License: MIT. Review its terms before using, modifying, or distributing the project.
• Unlocking the iPhone pauses mirroring, and a PIN-locked Android needs the user.
• OCR sees text rather than icons; unlabeled controls need a screenshot and a vision-capable model.
• Multi-touch, camera, Face ID, and DRM video are not supported.
• Connecting the phone is always the user's responsibility.
• Phone Harness Cloud is marked as an open beta.