SideTap
SideTap is an open-source MIT tool that lets an LLM agent see and control a real iPhone from a Windows desktop over USB. It does not require a Mac, Xcode, jailbreak, Appium server, or paid Apple developer account; a free Apple ID is enough. The install script sets up Python, go-ios, and SideTap on Windows, then the viewer walks through one-time iOS Developer Mode and WebDriverAgent sideloading steps. SideTap drives the phone through WebDriverAgent, so agents read the actual UI element tree with exact buttons, labels, and point coordinates rather than relying on OCR guesses. A Python harness exposes helper calls for tap, swipe, type, screen reads, and one-call flows like send_message, read_messages, and send_image. It also provides native MCP tools for agents such as Claude Code and Claude, plus a live viewer at around 34 fps with click-to-tap and drag-to-swipe controls. Safety and supervision features include a kill switch, live activity feed, and a prompt injection gate that pauses outgoing sends for approval after the agent has read the screen or messages. The gate bounds what an injected instruction can send, though the README states it does not bound what the agent can tap. SideTap also supports copying new camera roll photos and videos to a Windows folder over USB without iCloud.
- Category
- LLM
- Pricing
- Unknown
Problem solved
iPhone automation traditionally assumes the user owns a Mac and uses Xcode, WebDriverAgent, Appium, a paid Apple developer account, or a jailbreak. SideTap solves this by controlling a real iPhone from Windows over USB with a free Apple ID, giving LLM agents direct access to the iOS UI element tree for precise taps, swipes, typing, and screen reads.
Use cases
- Control a real iPhone from a Windows PC using an LLM agent over USB
- Automate iOS app UI interactions and reads without a Mac or Appium server
- Send and read iMessages or SMS through one-call agent flows from Windows
- Copy new iPhone camera roll photos and videos to a Windows folder over USB
- Register SideTap's helper API as native MCP tools in Claude Code or Claude
- Supervise agent phone actions with a live viewer, kill switch, and activity feed
- Approve agent-generated outgoing texts through the prompt injection gate
- Use precise UI element queries such as tap_text and scroll_until_found instead of OCR
Audience
- Developers
- ML Engineers
- Hobbyists
- Researchers
Pros
- Controls a real iPhone from Windows without requiring a Mac, jailbreak, or Appium server
- Uses WebDriverAgent to read the actual iOS UI element tree for exact taps and screen reads, not OCR guesses
- Offers native MCP tools for LLM agents, including Claude Code and Claude, plus a live viewer
- Includes safety controls such as a kill switch, live activity feed, and prompt injection gate for outgoing sends
- Open source under MIT with a one-line PowerShell installer and camera roll copy over USB
Cons
- Requires one-time iOS Developer Mode and WebDriverAgent sideloading with a free Apple ID, roughly 10 minutes of Apple steps
- Needs Windows 10/11, Python 3.10+, Node.js, and a USB-connected iPhone running iOS 17 or later
- Prompt injection gate bounds what an injected instruction can send, but not what an agent can tap or interact with
- Depends on a USB cable and WebDriverAgent; setup issues may require running phone-harness doctor to repair
- No wireless operation is described on the homepage, and no native Mac support is provided
Official website: https://sidetap.io/
Canonical LambdaBase page: https://www.lambdabase.com/tools/sidetap