Icemoon ← Home

AI copilot

Control a real iPhone with AI — the Icemoon MCP server

Last updated: July 16, 2026

Icemoon gives AI assistants hands. Its built-in MCP server exposes every iPhone connected to your Mac as a set of tools Claude can call: read the screen, tap, swipe, type on the real keyboard, navigate apps, run batches, schedule jobs. You describe the goal in chat — the AI plans and executes it on the physical device, and you watch the plan progress live on your dashboard. Stock iOS, no jailbreak.

What is MCP?

The Model Context Protocol is an open standard, introduced by Anthropic, that lets AI assistants call external tools and data sources. A weather MCP server lets the AI check the forecast; Icemoon's MCP server lets it operate a real iPhone. Anything you could do with your thumb, the AI can now do on your behalf — with the same human-like touch synthesis Icemoon uses everywhere: unique Bezier-curve swipes, natural timing, typing on the real iOS keyboard.

1

Connect your iPhone

Icemoon runs on your Mac; iPhones connect over USB. The one-time setup installs a signed automation runner on the device — no jailbreak, the phone stays on stock iOS.

2

The MCP server registers itself

Icemoon configures Claude Desktop automatically — open Claude and the device tools are simply there. Any other MCP-compatible client (Claude Code, Cursor, and others) can connect to the same local server.

3

Describe the goal

The AI reads the screen, identifies where it is, builds a step-by-step plan, and executes it — while the plan and each step's status stream live to your dashboard. If the app changes mid-task, the AI sees the new screen and adapts.

What the AI can do

CategoryToolsWhat it means
See the screenget_visible_elements · identify_screen · get_screen_info · screenshot_saveThe AI reads real UI elements — buttons, labels, fields — not just pixels, so it understands what it's looking at.
Touch & typetap · swipe · double_tap · touch_hold · pinch · scroll · type_text · smart_type · press_buttonEvery gesture is synthesized human-like: unique curves, natural timing, the real iOS keyboard.
Navigateopen_app · navigate_to · wait_for_screen · get_app_listOne call takes the AI to a target screen: it detects where it is, finds the path, walks it, and verifies arrival.
Go fastrun_batch · plan_create · plan_update_stepMulti-step sequences execute in one round-trip; plans stream progress to your dashboard as they run.
Device controllock_device · unlock_device · pasteboard_set · video_record · set_locationLock, unlock, clipboard, screen recording, and simulated GPS location for testing location features.
Orchestratelist_devices · connect_device · schedule_task · list_saved_scriptsPooled connections across the fleet — the AI switches devices instantly, runs saved automations, and schedules future jobs.

Things you can just ask

Open Settings, turn on Do Not Disturb, then verify it actually stuck.
Go through the onboarding of the app I just installed and take a screenshot of every screen.
Run the saved “morning report” automation on all connected iPhones and tell me if any device fails.
Navigate to the checkout screen in our app on every device and record a video of the flow.
Every day at 9:00, open the app, refresh the feed, and save a screenshot — schedule it.
Local by design. The MCP server runs on your Mac and devices connect over USB — Icemoon sends nothing to its own servers. Screen content reaches only the AI assistant you connect, and only while it's working for you.

Why a real device matters

Simulators don't run most real-world apps, don't have real cameras or push tokens, and behave differently under load. An AI agent that operates an actual iPhone works with the truth: the app store build, the real keyboard, the actual network conditions. For testing, that's fidelity; for daily workflows, it's the only option — your accounts and apps live on real hardware.

Get started

Setup takes about 15 minutes end to end: install Icemoon, connect an iPhone over USB, open Claude. The AI copilot and MCP server are included in the Farm and Scale plans.

See pricing Installation guide

FAQ

What is an MCP server for iOS?

MCP (Model Context Protocol) is an open standard that lets AI assistants like Claude call external tools. Icemoon's MCP server turns real iPhones into such tools: the AI can read the device screen, tap, swipe, type, navigate apps, and schedule tasks — on stock iOS, no jailbreak.

Which AI assistants work with Icemoon?

Icemoon registers itself with Claude Desktop automatically on setup. Because MCP is an open standard, any MCP-compatible client — Claude Code, Cursor, and others — can connect to the same server.

Does the AI see my phone's data? Where does it run?

The MCP server runs locally on your Mac, and devices connect over USB — Icemoon sends nothing to its own servers. Screen content is shared with the AI assistant you connect (for example Claude) only while you're actively giving it tasks, under that assistant's privacy terms.

Can the AI control several iPhones at once?

Yes. Device connections are pooled: the AI can switch between connected iPhones instantly or target a specific device for a single action, and you watch progress for the whole fleet on the dashboard.

Do I need to write code to use the AI copilot?

No. You describe the goal in plain language in the chat. The AI reads the screen, builds a step-by-step plan (visible live on your dashboard), and executes it. Anything it does can also be recorded and saved as a repeatable automation.

Which plans include the MCP server?

The AI copilot and MCP server are available on the Farm and Scale plans. Solo covers the visual flow builder, action recording, and the scheduler for a single device — see pricing for the full breakdown.

vs Appium vs AutoTouch vs Apple Shortcuts Install guide Home