# What is a cloud phone for AI agents?

A phone in a data center that your agent drives over an API or MCP. How it differs from a device farm and a multi-account phone, and five questions to choose one.

Decisions3 min read

In this article

A cloud phone for AI agents is a phone that runs in a data center and that an agent controls over an API or MCP: it reads the screen as text, taps, types and installs apps, and code starts, stops or deletes it. It gives an agent the apps that have no website, and gives a coding agent a place to run the app it built without a device on your desk.

Sources reviewed on November 9, 2026: [Mobilerun pricing](https://mobilerun.ai/pricing/), [Firebase Test Lab pricing](https://firebase.google.com/docs/test-lab/usage-quotas-pricing), [GeeLark](https://www.geelark.com/pricing/) and [mobile-mcp](https://github.com/mobile-next/mobile-mcp). Mobilebox facts come from the [API reference](https://mobilebox.dev/docs/reference).

## Three things are called a cloud phone

| Kind | Built for | How it is driven | Example |
| --- | --- | --- | --- |
| Agent cloud phone | An agent that acts in apps, step by step | REST or a hosted MCP server: read the screen, act by number | Mobilebox, Mobilerun, phonebox.dev |
| Device farm | Test suites across many devices | A test framework, run by run | Firebase Test Lab: $1 an hour virtual, $5 physical |
| Multi-account phone | People running many app accounts | A dashboard and scripts | GeeLark: "Your AI copilot for social media marketing" |

Only the first kind is made for an agent that decides its next step from what it sees. A device farm runs a test you wrote in advance and returns the result. A multi-account phone is made for a person at a dashboard.

## What the agent actually does

An agent cloud phone turns a phone into a loop of tool calls. On Mobilebox:

```
create_phone → observe → act → observe → park_phone
```

`observe` returns the screen as numbered elements, one per line, such as `[3] Text "Search" (540,288)`. `act` takes a list of taps, typing and swipes by number and returns the new screen. Your agent’s own model chooses each step; the phone runs no model of its own. The [agent guide](https://mobilebox.dev/docs/agents) walks through the loop, and the [live view](https://mobilebox.dev/live-view) lets you watch it.

## Cloud phone, local emulator or your own phone?

- **A local emulator with mobile-mcp:** free and fast on a laptop with hardware acceleration, but it lives on that laptop, and it stalls in cloud agent sandboxes and CI runners without KVM. See [cloud phone or local emulator](https://mobilebox.dev/blog/mobilebox-vs-local-emulator).

- **Your own phone over USB:** real hardware and your own accounts, one phone, on your desk.

- **A cloud phone:** reachable from anywhere an agent runs, one per task if you want, and kept between runs when you park it.

## How to choose one

1. **Where does your agent run?** A cloud agent or a CI job needs a URL and an API key, not a USB cable.

2. **Android, iPhone or a real device?** Apps that check for real hardware need a physical phone; see [real devices or cloud phones](https://mobilebox.dev/blog/real-android-devices-vs-cloud-phones).

3. **Does it keep logins?** Mobilebox keeps apps, data and logins when you park a phone. Mobilerun says "Stopped phones keep apps and data and bill no minutes."

4. **What do idle and boot time cost?** Mobilebox bills $0.01 per ready phone-minute, per second, with boot time and parked phones free. Mobilerun charges $0.03 per extra minute on its paid plans.

5. **How fast does it start?** We measured from under a second to 23 seconds from create to first screenshot across three services; see the [benchmarks](https://mobilebox.dev/benchmarks).

Six agent cloud phones side by side: [cloud Android phones for AI agents, compared](https://mobilebox.dev/blog/cloud-android-for-ai-agents-compared). Which one fits which job, with measured start times: [the best cloud phones for AI agents in 2026](https://mobilebox.dev/blog/best-cloud-phones-for-ai-agents). Mobilebox calls its phones [mobile sandboxes](https://mobilebox.dev/mobile-sandbox): a cloud phone your agent can use and throw away.

## Try it in your agent in 60 seconds

Paste this into Claude Code, Cursor, Codex or any agent that can read a web page. It asks for your email, gets its own key at once and adds the phone tools; the 6-digit code from your inbox keeps the workspace.

```
Set up Mobilebox for me: read https://mobilebox.dev/setup.md?via=what-is-a-cloud-phone-for-ai-agents and follow it.
```

[What your agent reads](https://mobilebox.dev/setup.md?via=what-is-a-cloud-phone-for-ai-agents)

In this article

[All articles](https://mobilebox.dev/blog)

## Next questions

- [How do I add it to Cursor, Codex or VS Code?](https://mobilebox.dev/blog/android-mcp-cursor-codex-vscode?via=next-setup)

- [Can my agent run 50 phones in parallel?](https://mobilebox.dev/blog/run-android-phones-in-parallel?via=next-parallel)

- [Can my agent use an iPhone?](https://mobilebox.dev/ios?via=next-iphone)

- [Can each phone get a residential IP?](https://mobilebox.dev/blog/residential-ip-per-phone?via=next-residential-ip)

- [Does my agent stay logged in between runs?](https://mobilebox.dev/blog/park-resume-android?via=next-logins)

- [Can I watch my agent work live?](https://mobilebox.dev/live-view?via=next-live)

- [What does it cost for a team?](https://mobilebox.dev/blog/mobilebox-pricing-for-teams?via=next-team)

- [How fast do cloud phones start, measured?](https://mobilebox.dev/benchmarks?via=next-benchmarks)