Claude Code now drives the iOS Simulator directly

The gap in agentic coding has always been the last mile: the agent writes the screen, then somebody has to open it and look. Anthropic closed that for iOS. Claude Code Desktop now opens a live iOS Simulator pane beside the conversation — Claude builds the app, installs it, taps through it, reads the screen, and fixes what it finds. It's in public beta on macOS, and the way it gets access is more interesting than the feature itself.

What actually happened

The simulator pane drives Apple's simulator directly rather than going through computer use. That's the design decision worth noting: it never takes over your screen or hides your windows, and it doesn't need the macOS Accessibility and Screen Recording permissions computer use requires. From the CLI, Claude still reaches the simulator the old way — by controlling it on your screen like a person would.

Requirements are narrow: Claude Desktop v1.24012.0 or later, a Mac, and Xcode with the iOS platform installed. It's on Pro, Max, and Team plans — not Enterprise. Local sessions only; cloud and SSH sessions run on a machine that can't reach simulators on your Mac. Each session owns its devices, up to four panes, and Claude shuts down the simulators it booted ten minutes after you detach one. The pane is interactive both ways — you can tap and swipe while Claude works, though a "Claude is using this device" badge tells you when to keep your hands off.

Two things follow your permission mode rather than the one-time device consent: building the app, because xcodebuild runs your project's build scripts on your Mac, and opening a URL on the device, because a URL can carry data off it. Organizations can kill the whole thing with the disableMobileSimulatorTools managed setting. Claude can't touch a physical iPhone — simulators only.

Why this matters for your business

The real unlock is the verification loop. An agent that can only write code has to hand you a diff and hope. An agent that can run the app, tap the signup flow, and read what came back can catch its own mistakes before you ever see them. That's the difference between AI that generates work and AI that finishes it — and it's the same reason we build automations with a check step instead of a confidence score.

Now read the privacy line, because Anthropic put it in the docs and most people will scroll past it: Claude's screenshots of the device are sent to Anthropic and kept under your normal conversation retention settings, and the docs explicitly say not to sign in to real accounts on a device Claude uses. For an app with a login screen, that's a policy decision, not a preference. Use seeded test accounts. If you're under a client NDA or handling regulated data, decide before your first session whether simulator screenshots are allowed to leave the machine — and if they aren't, disableMobileSimulatorTools exists for exactly that.

Sources: Claude Code Docs, MacRumors.