← all repositories
JosssphZhou/poki

An AI agent that borrows your actual iPhone, no jailbreak

Poki is a local MCP server plus agent that taps, swipes, and types on a real iPhone from your Mac — nothing installed on the phone, nothing cracked open.

★611 stars JavaScript Coding AssistantsAgents
poki
Collecting fresh signals — velocity needs a few days of history.
collecting data…
star history

What it does

Poki exposes your iPhone as a set of MCP tools, so an agent (Claude Code, Codex, or Poki’s own terminal agent) can drive a real device: tap, swipe, type, read the screen, and run multi-step tasks like “search esp32 in Chrome and summarize the top result.” It works against a USB-paired physical iPhone or a simulator, and the phone stays stock — no jailbreak, no extra hardware. There’s also a standalone agent with iMessage and Telegram channels, so you can text your phone instructions.

The interesting bit

The agent remembers past tasks and re-runs them faster on repeat — self-learned routes, not just blind replay. And the community has pushed it in directions the author didn’t plan: it runs on a router so Telegram can drive the iPhone with the Mac powered off, and there are Linux and Windows ports.

Key highlights

  • Local MCP server; works with any client supporting mcpServers JSON, using your existing model subscription
  • Physical iPhone over USB/Wi-Fi (Tailscale remote connections supported), or iOS simulator
  • iMessage and Telegram channels for messaging-driven control
  • Self-learning routes speed up repeated tasks
  • AGPL-3.0, macOS-focused with community Linux/Windows ports

Caveats

  • Heavy prerequisites: macOS 15+, full Xcode, Node.js 24+, Rust, ffmpeg — the prebuilt package trims some of that but still needs Xcode
  • Physical-device verification has only been done on iPhones with the Chinese interface; portrait orientation only, and some features (app container export, direct Settings deep links) are unavailable
  • It controls your real phone: ask it to send messages or make payments and it will, and screen contents go to your configured model service

Verdict

Worth a look if you want an agent that operates iOS the way a human does — especially for repetitive phone chores. Skip it if you’re not on a Mac with Xcode or if a simulator-only workflow would suffice with simpler tooling.

Frequently asked

What is JosssphZhou/poki?
Poki is a local MCP server plus agent that taps, swipes, and types on a real iPhone from your Mac — nothing installed on the phone, nothing cracked open.
Is poki open source?
Yes — JosssphZhou/poki is open source, released under the AGPL-3.0 license.
What language is poki written in?
JosssphZhou/poki is primarily written in JavaScript.
How popular is poki?
JosssphZhou/poki has 611 stars on GitHub.
Where can I find poki?
JosssphZhou/poki is on GitHub at https://github.com/JosssphZhou/poki.

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.