Claude Code Now Tests iOS Apps Independently: It Builds, Interacts, and Iterates Until They Run

Anthropic has updated Claude Code, the AI-assisted development environment, to communicate with Apple's iOS simulator: starting yesterday, the integration is available in public beta on the desktop version for Mac. When asked to compile, launch, or verify an app, it appears in a dedicated panel next to the conversation, where the code runs live.

The behavior mimics that of a developer at work. Claude installs the app on the simulated device, launches it, interacts with the screens to check its modifications, then corrects and retries until the results hold up. The suggested example in the documentation is building an app and running it in the simulator to verify the onboarding procedure. The panel remains interactive for onlookers, who can touch, drag, press hardware buttons, rotate the device, or save screenshots and recordings while the assistant continues to work. A single session can manage up to four devices in parallel.

Why Screen Capture Is Not Needed

The panel directly drives the simulator, so Claude does not resort to the so-called computer use mode, the one where it moves the pointer and clicks on the screen like a user would. As a result, the integration does not require macOS Accessibility and Screen Recording permissions, and the assistant does not take control of the monitor nor hide other windows. In contrast, from the command line, Claude continues to operate in computer use mode, with visible control of the simulator on screen.

The requirements remain tied to the Apple environment. You need Claude Desktop version 1.24012.0 or later, a Mac (the iOS simulator only runs on macOS), and Xcode with the iOS platform installed, which provides the virtual devices. The function operates only in local sessions: those in the cloud or via SSH run on a machine that does not reach the simulators on the user's Mac. Access is limited to Pro, Max, and Team plans.

The screenshots that Claude captures from the device are sent to Anthropic and stored according to the normal conversation retention settings; therefore, the company recommends not to log in with real accounts on a device used by the assistant. The first use of each simulator requires explicit consent, valid once per device, while those managing Claude Code can entirely deactivate the simulator tools through a managed setting.