Guide
How to let an AI agent use a real iPhone
Published October 7, 2026
Yes, an AI agent can use a real iPhone, with a Mac in the loop. The agent sees the phone's screen, then taps, swipes and types the way a person would, and checks the result before its next step. That matters because plenty of tasks don't live in a web browser: app-only services, mobile-only flows and Messages.
This guide explains how it works, compares the two practical routes (build it yourself with an open-source MCP server, or use Fonino), and is honest about what a real phone won't fix.
Can an AI agent really use an iPhone?
Yes. iOS doesn't let one app control other apps, so the agent works through a Mac that drives the phone with Apple's XCTest framework, usually through a helper app called WebDriverAgent. The agent sees the screen, then taps, swipes and types, and checks what changed.
Why it needs a Mac
Apple only allows input to be injected into a physical iPhone by a signed test runner (XCTest) installed on the device, and that runner is built and signed with Xcode on a Mac. The Mac then relays the agent's actions to the phone over USB. Android is different: it ships a debug bridge (ADB) for input and screen dumps, which is why phone-control tools appeared there first.
What about the assistant apps on the phone?
Assistant apps installed on an iPhone can use some iOS features through system permissions, like starting a message or adding a calendar event. They can't operate other apps the way a person does. For full control of a physical iPhone, the agent needs a tool that drives the device.
What an agent can do on a real iPhone
- Use native apps: open an app, tap, swipe and type, then look at the screen again to confirm.
- Browse on the phone: use Safari on the device, with the phone's own signed-in sessions.
- Use Messages: read a conversation and send texts from the phone's own number.
A simple example from Fonino's demo: “Open Notes and write a short note.” The agent opens Notes, taps the compose button, types the note and checks that it saved. Whether a specific third-party app works depends on the app, its sign-in requirements and the agent.
Option 1: Do it yourself with an open-source iPhone MCP server
An iPhone MCP server exposes phone actions (screenshot, read the screen, tap, swipe, type, launch apps) to any agent that speaks MCP, the open standard for agent tools. Good open-source options exist for developers:
- iPhone-mcp: an MCP server for a physical iPhone, with a live viewer.
- ios-mcp: a Python MCP server with an approval gate and a clear list of what it doesn't do.
- mobile-mcp: one MCP server for iOS and Android devices and simulators.
- Appium MCP: MCP on top of Appium, the long-standing mobile test framework.
The typical steps
- Install Xcode on a Mac and add your Apple ID.
- Turn on Developer Mode on the iPhone and trust the Mac.
- Build and sign WebDriverAgent with your Apple developer identity, and install it on the phone.
- On some tools, set up a tunnel or port forwarding to reach the phone.
- Add the MCP server to your agent's configuration.
Where people get stuck
The signing step is the usual wall. With a free Apple ID, Apple's signature for the helper expires after 7 days, so DIY setups have to be rebuilt and re-signed by hand every week, or you pay $99/year for the Apple Developer Program. Tunnels and MCP configuration files are a second hurdle for anyone who isn't a developer. For a developer's first-hand account, see Driving a Real iPhone from Claude Code.
Option 2: Use Fonino, with your own iPhone or a hosted one
Your own iPhone (BYOF)
Use a spare iPhone and a Mac with Apple silicon, connected by USB. The Fonino app does the rest in about 10 minutes.
- No Apple developer account. Fonino signs its phone helper for you.
- No re-signing. Fonino keeps the phone helper signed for you. More in the FAQ.
- Price: 14 days free with just your email, then $9/month founding price for the first 100 members (regular price $19).
A hosted iPhone
No spare iPhone? Fonino keeps a dedicated iPhone in its rack, on Wi-Fi, charged and updated. You sign it in with your Apple Account. It texts through iMessage over Wi-Fi, with an optional eSIM of your own for SMS and a number; it doesn't place calls. $149/month billed annually ($1,788/year), with a 30-day money-back guarantee.
Set it up from inside your assistant
Paste this into your assistant and it walks you through each step: what to click, type or tap, and when you're done.
Start in your assistant. Paste this into Grok Bot or Muse and follow along:
Set me up with Fonino so you can use a real iPhone for me. Read https://api.fonino.app/setup.md and walk me through it one step at a time: tell me exactly what to click, type or tap, and wait until I say done before the next step.Have a spare iPhone and a Mac? Your assistant sets you up with BYOF. No spare iPhone? It points you to a hosted phone we run for you. You’ll need a Mac (Apple silicon) and your iPhone plugged into it by USB. About 10 minutes. Your assistant signs you up (free trial) and walks you through installing the Fonino app on your Mac, which sets up your iPhone. Using Claude Code or Codex? Paste it there and it does the steps for you.
Using Claude Code or Codex on your Mac? Paste it there and it runs the steps for you. Fonino works with any agent that can add a remote MCP server; see the documentation and the Hermes guide.
Staying in control: approvals, blocked apps, live view
A tool that can tap anything on your phone needs guardrails. Look for approval before anything is sent, posted or bought; blocked sensitive apps; a live view; and a way to take over. Fonino does this:
- One-tap approval. Posts, replies, texts and purchases wait for your OK, showing the app, the destination, the exact text or items, and the phone's screen. One approval covers one action, and the assistant can't approve itself. No answer within 10 minutes means nothing happens.
- It checks the button under every tap. Buttons like Send, Post, Place your order and Buy Now need your approval even if the assistant doesn't ask.
- Sensitive apps are blocked: Settings, Wallet, Passwords, the App Store, Mail, Phone, FaceTime and Find My.
- Logins stay with you. At passwords and 2-factor codes, the assistant hands the phone to you with a private live view, and one-time codes in texts are hidden from it.
- Watch and take over at any time, and disconnect an assistant with one click. With BYOF, the phone and its recordings stay on your Mac, with no open ports.
More detail on the security page.
What a real iPhone won't fix
- Apps and websites can still ask for verification or restrict access. Plan for a person to step in at sign-ins, CAPTCHAs and 2-factor codes (how to hand those off).
- Not every app or workflow has been tested.
- With your own iPhone, sending through iMessage has been tested; SMS still needs validation.
- A hosted phone can't reach Android users over iMessage and doesn't place calls. Live voice calls are in development.
DIY vs Fonino: which should you choose?
| DIY open-source MCP server | Fonino (your own iPhone) | |
|---|---|---|
| Apple account | A developer identity to sign WebDriverAgent; free Apple IDs work but signatures last 7 days ($99/year to avoid that) | A free Apple ID works; no paid developer account |
| 7-day re-signing | Manual: rebuild and re-sign when it expires | Automatic, in the background, while the iPhone is connected |
| Setup | Xcode, Developer Mode, WebDriverAgent build and signing, sometimes tunnels, plus MCP configuration | Xcode download, installer and guided setup page (about 20 minutes plus Xcode) |
| Hardware | Mac and iPhone | Apple-silicon Mac and iPhone on USB, or a hosted phone with no Mac |
| Guardrails | Varies by tool | One-tap approval with the exact text or items; sensitive apps blocked |
| Live view and takeover | Some tools | Yes |
| Cost | Free software, plus your time | $9/month founding price (first 100), then $19; hosted $149/month billed annually |
Build it yourself if you're a developer who wants full control and enjoys tinkering; the open-source tools are good. Use Fonino if you'd rather skip developer accounts and weekly re-signing, and want guided setup, approvals and a hosted option.
FAQ
Can an AI agent use a real iPhone?
Yes. A Mac drives the phone through Apple's XCTest framework, and the agent connects through a tool such as an MCP server. With Fonino, your assistant sees the screen and taps, swipes and types in native apps, Safari and Messages.
Do I need a Mac?
To use your own iPhone, yes: a Mac with Apple silicon, with the iPhone connected by USB. If you don't have a spare iPhone and Mac, Fonino's hosted phone runs in Fonino's rack, and you sign it in with your Apple Account.
Do I need a paid Apple developer account, and do I have to re-sign every week?
No and no. A free Apple ID works. Apple's signatures for free accounts expire every 7 days, so Fonino renews them automatically in the background while your iPhone is connected.
Which AI assistants work with Fonino?
Grok Bot, Muse and Hermes Agent have been tested with Fonino on physical iPhones. Any agent that can add a remote MCP server can connect. Compatibility with individual apps and workflows varies.
Will my agent send messages or buy things without asking?
No. Posts, replies, texts and purchases wait for your one-tap approval showing the exact text or items. One approval covers one action, and if you don't answer within 10 minutes, nothing happens.
Does a real iPhone prevent blocks or CAPTCHAs?
No. Apps and websites can still apply their own checks. At passwords or 2-factor codes, the assistant hands the phone to you.
How much does it cost?
Using your own iPhone is free for 14 days with just your email, then $9/month founding price for the first 100 members (regular price $19). A hosted phone is $149/month billed annually ($1,788/year), with a 30-day money-back guarantee.
Can it make phone calls?
Not yet. Live voice calls are in development. The hosted phone texts through iMessage over Wi-Fi and doesn't place calls.
Give your assistant its own iPhone
Start free for 14 days with just your email. Paste one message into your assistant and it sets you up.