AGENTIC COMMONSAI industry briefings

繁中EN

TopicAI Coding & Developer ToolsPublished

All English articles

Claude Code Desktop Boots a Live iOS Device Pane

Summary: Claude Code Desktop now ships a public beta iOS Simulator pane on macOS: when Claude builds, installs, launches, or checks your iOS app, the pane opens next to the conversation and streams the simulated device live. Claude installs the app, taps through it, and reads the screen to verify its own changes while you can tap the same device. For iOS builders on Anthropic’s paid plans, the loop between a code change and seeing it on screen just got shorter.

How the pane works

There is no command to run. Open your iOS project in the desktop app, ask Claude something like “build the app and run it in the simulator to check the onboarding flow,” and the pane appears once the app launches. Under the hood, the pane drives Apple’s simulator tooling directly — which the desktop app doesn’t include, so you need Claude Desktop v1.24012.0 or later, a Mac, and Xcode with the iOS platform installed (Xcode 26.x or 27 recommended).

The direct connection matters more than it sounds. From the CLI, Claude reaches the simulator through computer use, controlling it on your screen the way you would with a mouse. The pane skips that entirely: it never takes over your screen, hides your windows, or needs the macOS Accessibility and Screen Recording permissions computer use requires. You and Claude share the device — your taps change the app state Claude sees — and a “Claude is using this device” badge tells you when to keep your hands off.

The first time Claude touches a simulator, you allow it once per device. After that, tapping, typing, and screenshots run without prompts, but three actions still follow your session’s permission mode: opening a URL on the device (a URL can carry data off it), building the app (xcodebuild runs your project’s build scripts), and installing and launching the app (Claude can run any app it names, not just yours).

The privacy catch is concrete: Claude’s screenshots of the device are sent to Anthropic and kept under your normal conversation retention settings. The docs’ advice is blunt — don’t sign in to real accounts on a device Claude uses. Organizations get two off switches: disableMobileSimulatorTools blocks Claude’s tools while leaving the pane usable for your own taps, and requireCoworkFullVmSandbox runs tools in an isolated VM and disables the pane entirely.

Where it stops

Three limits decide adoption for many teams. First, simulated devices only — Claude cannot control a physical iPhone or iPad, so hardware-dependent behavior still routes through Xcode and screenshots you attach yourself. Second, the pane works in local sessions only; in cloud or SSH sessions, Claude runs on a machine that can’t reach your Mac’s simulators. Third, devices Claude booted shut down when you quit the app, when you archive the session, or 10 minutes after you detach. If you want the foldable path, the iPhone Duo needs Desktop v4.29.0 or later plus Xcode 27.1 with its iOS simulator runtime, and hinge angles from 1 to 179 degrees. The docs page goes further on Duo details and troubleshooting than this article covers.

It’s also worth noting that Claude Code already reaches your phone from another direction, via the remote control feature — this pane is the app side of that mobile story.

Our take: If you already pay for Claude and build iOS on a Mac, act on this now: it’s a shipped beta, the experiment costs an afternoon, and turning it off is one settings change. The reason to move isn’t a benchmark — Anthropic publishes no accuracy numbers for how reliably Claude verifies app behavior on screen, so we’re judging the mechanism, not a measured result. Treat the simulator as a synthetic-data environment: no real accounts, no production sign-ins, because every screenshot leaves your machine. If you’re on a third-party provider, a non-macOS setup, or cloud sessions, you can safely ignore this — it simply won’t attach. Teams whose critical flows depend on real hardware should wait too; the pane adds nothing there. What would reverse our recommendation: evidence that Claude mis-reads simulated screens often enough that its “verified” claims need re-checking by hand, which would reduce the loop to a screen mirror you drive yourself.

Sources

Original post
AGENTIC COMMONSOperated by PHLEGON LABS
ShareXEmail
Support us

Related reading

  1. Claude 5 context engineering and system prompt rules

    Migrating teams should trim selectively: Anthropic cut over 80% of Claude Code's prompt without measurable loss on coding evaluations.

    Claude

  2. Visa and Replit agentic payments integration

    Replit enterprises can assess its self-serve tier up to $200,000; agent payment builders should await integration details.

    Agentic AI

  3. Claude Code Mods Run Your Code Inside the Agent

    Claude Code mods let you theme the desktop app with JavaScript plugins that run unsandboxed with your own permissions.

    Claude Code