ESP32 iPhone Agent MCP
MCP server and plugin: an agent sees the iPhone screen through a $7 ESP32 board, then taps, swipes and types with no iOS changes.
- 4,907 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- The agent acts on your behalf on a phone with access to signed-in accounts and apps.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/esp32-iphone-agentDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The agent acts on your behalf on a phone with access to signed-in accounts and apps.
This is third-party code. Review the repository files before installing.
What it does
ESP32 iPhone Agent connects an iPhone to an AI model through an ESP32 board. The board receives AirPlay screen mirroring and acts as a USB mouse and keyboard, so the agent sees the screen, taps, swipes and types text. The agent can run inside an app on the phone with no computer. For coding agents there is the `esp32agent` npm package, run as an MCP server with `npx -y esp32agent mcp`. The repository also works as a Claude Code plugin with a `handsy` skill. The board is reached over Bluetooth or a shared Wi-Fi network; board firmware 0.2.2 or newer is required.
Who it is for. Mobile developers and testers who want an agent to drive a real iPhone without a jailbreak or a simulator.
Good fit when
- A scenario must run on a real iPhone because the simulator behaves differently.
- An agent should set an alarm or change settings through the phone interface.
- Saved tasks are triggered from Siri, Shortcuts or a widget.
- An app is checked step by step and some steps go to an agent.
Not a fit when
- You need passcode, Apple Pay or DRM video screens, which iOS does not mirror.
- The app requires landscape orientation.
- You have no right to control the phone and the accounts signed in on it.
Example request
On the iPhone open Clock, set an alarm for 8:00 on Mondays and Tuesdays and save it.MCP
- Transport
- stdio
- Authentication
- not required
Limitations
An ESP32 board with firmware is required. Mirroring works only from an iPhone, not macOS, at 480 by 1040 and up to 5 frames per second; each step takes seconds. iOS hides passcode, Apple Pay and DRM screens, and mirroring stops on lock. While running, the phone Wi-Fi is the board hotspot and the model is reached over cellular. Chinese is typed as pinyin and emoji are dropped.
How to disable. Run `/plugin uninstall esp32agent@handsy` or delete the `esp32agent` entry from MCP settings. Unplug the board and stop any task from the Dynamic Island.
FAQ
How do I connect it to Claude Code?
Use `/plugin marketplace add onehumanbeing/esp32-iphone-agent` then `/plugin install esp32agent@handsy`. For the MCP server alone, run `claude mcp add esp32agent -- npx -y esp32agent mcp`.
Is a computer needed?
No, in the `main` version the agent runs in an iPhone app. A computer is needed only for the MCP mode and for the old `v0.1.0`.
Where do I get the app?
Through TestFlight from the README link; the App Store listing is in review. The board still has to be flashed per the quick start.
Which screens will the agent miss?
Passcode entry, Apple Pay and DRM video. iOS does not send them to mirroring.
What does the license require?
The project is MIT: keep the copyright notice and credit ESP32 iPhone Agent by Henry Yuan in your README, as the author asks.
Discussion
No questions or reviews yet.
Related
MCP serversAn app, CLI, MCP server and skill let Claude Code and Codex see and tap an unmodified iPhone, with no jailbreak and no cable.
An MCP server that installs on a jailbroken iPhone and gives an agent 46 tools: taps, screenshots, OCR, apps, files and logs
A plugin and MCP server that give an agent annotations on the iOS Simulator screen: element, screen, search strings, two shots.
Lets Claude Code and Codex drive a real iPhone: read the screen, tap, type and scroll in any app
iPhone automation on XCUITest with a JSON-RPC API: an agent taps, types and reads the screen from a manual served by the device.
62 MCP tools so an agent can tap, scroll and type on a real Android phone over ADB or on an iPhone.
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| ESP32 iPhone Agent (this card) | no | not required | 4,907 | |
| Mobile MCP | no | not required | 10,529 | an MCP for simulators and devices via system tools; this one drives an iPhone physically through an ESP32 board |
| iphone-use | no | not required | 494 | another iPhone MCP; ESP32 iPhone Agent can work without a computer or iOS changes |
README in short
The README explains the architecture, hardware, a quick start with flashing and TestFlight, connection from a coding agent, limitations and a responsible-use section.