What are cloud phones for AI agents?.

Key moments

  1. 0:00 What are cloud phones for AI agents?
  2. 0:18 Why do AI agents need a phone?
  3. 0:44 Cloud phone vs emulator, device farm or phone number
  4. 1:17 How does an agent drive a cloud phone?
  5. 1:45 What should you look for in a cloud phone?
  6. 2:08 Phonebox: the agent-first cloud phone

Transcript

0:00 Cloud phones for AI agents are real Android phones, hosted in the cloud, that an AI agent controls through an API or an MCP server.

0:09 The agent reads the screen, taps, types and installs apps, with no phone on your desk and no emulator on your machine.

0:18 So why would an agent need a phone? Three reasons.

0:22 First, apps with no API.

0:24 A lot of real work lives inside mobile apps that have no API and no website.

0:30 Second, testing your own app.

0:32 The agent installs the APK it just built, and walks through the screens it changed.

0:37 Third, accounts that stay signed in.

0:39 Sign in once, and the phone keeps the account for the next task.

0:44 How is that different from what you might use today?

0:47 A local emulator with ADB runs on your own machine.

0:50 It needs the Android SDK, and it's hard to run in CI or in a cloud sandbox.

0:57 A device farm rents real devices for scripted test runs, and usually resets them after each session.

1:03 And a virtual phone number gets you calls and texts.

1:05 But a number is not a phone: there's no screen, and no apps.

1:10 A cloud phone for agents is a full Android device, on demand, from wherever your agent runs.

1:17 So how does an agent drive one?

1:19 It's a loop: observe, act, check.

1:23 Observe returns the screen as text: a short, numbered list of every button and field, with its label, plus a screenshot.

1:31 Act taps and types.

1:32 Orders. Reorder.

1:33 Six crates. Place order.

1:36 Then it observes again, to check the result. Order placed.

1:40 And Phonebox never retries a write on its own, so a slow response can't turn into a double order.

1:45 What should you look for in a cloud phone? Four things.

1:49 A hosted MCP server, so any agent can connect with one URL.

1:53 A human handoff: a live view link, for one-time codes and payments.

1:59 Persistence: parked phones that keep their apps and sign-ins.

2:04 And per-second pricing, so you only pay while the phone runs.

2:08 Phonebox is the agent-first cloud phone.

2:10 Claude Code, Codex, Cursor or your own agent connect through MCP, a CLI, an SDK or REST.

2:19 It costs six cents a minute, billed per second.

2:22 Parked phones are free, and an idle phone parks itself after five minutes.

2:27 It's Android only, with no phone number or SMS.

2:30 New accounts get two dollars of free credit, with no card.

2:33 Phonebox.dev.

Try it yourself

Paste Read https://phonebox.dev/setup.md and set me up with Phonebox into Claude Code, Codex or Cursor. Then read give your coding agent an Android device, the Android MCP server guide, or cloud phones for AI agents.

Watch on YouTube.

The MCP tool calls and their results in chapter 4 come from a real run against phonebox.dev/mcp, and the phone footage is that phone's own screen recording. The tool-call window is a recreation that shows those calls. Dockside Supply is a sample app built for demos.

Get started with Phonebox — $2 in free credit, no card required.

The Phonebox team

← All guides