Anthropic is giving developers a new way to build and test iOS applications. The Claude Code desktop client now integrates directly with Apple's iOS Simulator, and the feature is available in public beta starting today.
When instructed to build, run, or check an app, Claude Code can open the project inside the iOS Simulator pane. The AI assistant watches the simulator run in real-time, interacts with the interface, and iterates on the code until the application functions as intended. Developers can also continue using the simulator manually while Claude works in the background, creating a collaborative loop where human and AI adjustments can happen concurrently.
This integration drives the simulator directly. It does not rely on the general "computer use" capability that requires macOS Accessibility and Screen Recording permissions. Because Claude has direct access to the simulator environment, it bypasses those system-level hurdles entirely, controlling the screen just as a human user would but without the privacy overhead that comes with granting an AI broad access to your entire operating system.
There are privacy considerations to keep in mind. Claude's screenshots of the simulated device are sent back to Anthropic and retained according to standard conversation retention policies. Anthropic explicitly advises users not to sign into real personal accounts on any device that Claude is actively controlling, a reasonable precaution given that the AI is essentially observing and interacting with a live screen.
The feature requires Claude Code Desktop running on macOS, with Xcode and the iOS platform installed. The simulator pane is restricted to local sessions only, so it cannot be used in remote or shared environments.
This move signals a broader shift in AI coding tools. Until now, most AI assistants operated purely as text generators, producing code snippets that the developer had to manually compile and test. By stepping into the simulator, Claude is moving from writing code to actively participating in the build, run, and verify loop. If this approach works well, it could compress the development cycle significantly, especially for developers who spend a large portion of their time toggling between an editor and a simulator window.
The integration also raises interesting questions about where AI coding tools are heading next. Direct simulator access is a natural step beyond text generation, but it stops short of deploying to physical devices. Apple's strict controls over hardware access mean that testing on an actual iPhone still requires a human developer. Still, closing the gap between writing code and seeing it run on a simulated device removes one of the biggest friction points in mobile development, and it gives Anthropic a concrete differentiator in the increasingly crowded AI coding assistant market.