Tapkit lets Claude tap and swipe your real iPhone

Getting an AI to tap buttons on your iPhone has always run into Apple's walled garden.

Normally, automating anything on iOS meant fighting with Appium, messing with simulators, or jailbreaking the device. A new Mac application called Tapkit is taking a radically simpler approach: just plug your phone into your Mac.

Tapkit turns a physical iPhone into an agent playground, letting models like Claude and Codex see your phone's screen and interact with it directly.

No jailbreak, no apps, just a cable

The most surprising part about Tapkit is the zero-install setup on the mobile side.

You do not need to install an app on your iPhone, there is no jailbreak required, and you do not even have to enable Developer Mode. The only real constraint is physical: your iPhone must be plugged into a Mac running Tapkit.

Once hooked up, you can control the device using Tapkit's built-in agent, Claude, Codex, or custom workflows via its REST API and Python SDK. It also works out of the box with any Model Context Protocol (MCP) client, treating the physical phone as a standard modular tool.

Vision-driven taps and live human takeover

The mechanic is straightforward: the agent looks at screenshots, decides what to do, and executes taps, swipes, or keystrokes.

After each action, it snaps another screenshot to verify the result before moving forward. Because it runs on an actual device, all your signed-in accounts, daily apps, files, and notification permissions are already there. That lets agents stitch together mobile workflows across apps that lack public APIs.

Users can watch the actions unfold in real time, inspect previous session logs, and manually take over whenever a two-factor code or login screen pops up.

Why it matters

For mobile developers and QA engineers, this offers an immediate way to run cross-app automated testing without wrestling with simulators. If you already use Claude or MCP workflows, plugging in a spare iPhone gives you an instant testing rig.

We think the clever part here is sidestepping Apple completely. Instead of trying to get an autonomous agent approved on the App Store, Tapkit treats the iPhone as an external peripheral driven from macOS. Connecting that directly into MCP makes the agent setup remarkably frictionless.

That said, the physical tether is a clear limitation. This is not a personal assistant that helps you on the subway; it is a desk-bound rig. Visual grounding through screenshot loops can also be brittle when apps lag or show unexpected popups. It is an ingenious bridge for today, but it remains a stopgap until platforms offer native agent APIs.