Getting Started

iphone-use vs other iPhone automation tools

Several projects let an agent or a test drive an iPhone. They differ in what they target, how they reach a real phone, and what they are for. This page compares facts, not rankings.

Checked 2026-10-27, from each project's own README. Projects change; follow the links for the current state.

At a glance

Project	Targets	Real iPhone path	Interfaces	Host needs	License
iphone-use	Real iPhone (iOS 15+); iOS Simulator (preview)	Its own XCTest runner, over USB or an encrypted Wi-Fi tunnel	MCP, HTTP API, CLI, browser page, iOS app	Mac, macOS 15+, Xcode	PolyForm Shield 1.0.0 (source-available)
agent-device	iOS, Android, HarmonyOS (simulators, emulators, physical devices), TV platforms, web, macOS, Linux	XCTest for iOS interactions and physical iOS	CLI, MCP server, typed Node.js API	Node.js 22.12+	MIT
Mobile MCP	iOS and Android: simulators, emulators, real devices; cloud devices via Mobile Next Cloud	Device on USB and trusted; built on mobilecli	MCP	Node.js 20+, Xcode command line tools, Android platform tools	Apache-2.0
Appium XCUITest driver	iOS, iPadOS, tvOS; watchOS (Simulator only)	Apple XCTest via WebDriverAgent	WebDriver (Appium)	Appium 3 (driver 10.0.0+)	Apache-2.0
blitzdotdev/iPhone-mcp	Real iPhones and simulators	Builds and installs WebDriverAgent on the phone; USB or same Wi-Fi	MCP, browser live view (physical iPhone)	macOS, Xcode, Node.js 18+, Homebrew	MIT
ios-agent	iOS Simulator or a real iPhone, cable or Wi-Fi	WebDriverAgent; go-ios for physical devices	Terminal app, MCP, Python library	macOS, Xcode 16.3+, Python 3.12+	MIT
Lakr233/iphone-mcp	A real iPhone by UDID	Appium with the XCUITest driver and WebDriverAgent	MCP over streamable HTTP	Python 3.8+, Node.js, Xcode	MIT

Notes from the same READMEs: WebDriverAgent's repository says development moved to the appium-ios monorepo. Last pushes seen on GitHub: blitzdotdev/iPhone-mcp on 2026-04-01, Lakr233/iphone-mcp on 2025-09-01.

What iphone-use focuses on

Your own real iPhone and the apps already on it, signed in: Health, banking, chat apps, your app under test. A daemon on your Mac hosts the phone.
Its own XCTest device runner, which replaced WebDriverAgent in v0.14.0 and keeps the WebDriverAgent HTTP API. Requests to it are signed per launch.
Explicit action outcomes: applied, not sent or unknown, with retry_safe; taps on covered elements are refused.
Deterministic flow replay from a reviewed flow registry, with no model in the loop.
People and agents on one phone: a browser page and an iOS app for remote control, one owner at a time.
The host is a Mac, and it drives iPhones (and iOS Simulators in preview), not Android devices.

Which one fits

📱 Everyday apps on your own iPhone

Reading data out of apps with no API, with you able to watch and take over: what iphone-use is built for.

🤝 iOS and Android from one tool

agent-device and Mobile MCP both cover iOS and Android, simulators and physical devices, and connect to device clouds.

🧪 A maintained WebDriver test suite

Appium with the XCUITest driver is the established WebDriver ecosystem for cross-platform end-to-end tests.

🐍 A Python agent with its own loop

ios-agent ships a terminal agent, an MCP server and a Python library, for the iOS Simulator and real iPhones.

A measured comparison

ℹ️ Maintainer measurement, not an independent benchmark

The iphone-use README reports, for the same iPhone 13 over USB (October 2026): reading the screen took 0.08–0.135 s with the iphone-use runner and about 0.7 s with agent-device; a tap plus waiting for the screen to settle took 1.0 s and 2.8–3.3 s. No other tool on this page was measured by the project. Speed claims need the same device, actions and settling conditions to compare.

Facts on this page can go stale as these projects ship. If something here is wrong, please open an issue. More on how iphone-use is built: How it works.