Give your coding agent an Android device.

An Android device for your agent

Some tasks happen inside an Android app. Your agent may need to walk through a new build, check a signed-in flow, or interact with a mobile screen that a web browser cannot reach.

Phonebox provides the device. The agent can read the screen, tap, type, install an APK and move files. A parked device keeps its apps, files and signed-in accounts, so you can resume it later.

This is Android device control. Phonebox does not include a phone number, voice calls or SMS. If a task has a suitable API or works in a browser, your agent should use that route when it makes more sense.

Connect your coding agent

Paste this into Claude Code, Codex or Cursor. You do not need an account before you start.

Read https://phonebox.dev/setup.md and set me up with Phonebox from here: get me signed in and ready, then find out what I want to use phones for and help me with that.

1. Let your agent read the setup guide. It installs the Phonebox CLI and starts the sign-in flow.
2. Open the sign-in link. Sign in with Google, or ask your agent for an email link. Check the code shown in the browser against the one your agent gave you, then connect.
3. Choose the task together. Setup confirms your project and balance. Your agent reads the Phonebox skill before driving a device.

The setup flow saves an agent key on the machine your agent runs on. You do not need to paste an API key into the chat. You can revoke the key from the console.

Give it a small first task

Begin with a task whose result you can recognize. For an Android app you own, that might be installing a build and checking its welcome screen.

Use Phonebox to install ./app.apk and check the welcome screen.
Read the screen before each action and confirm the result.
Ask me before signing in or submitting anything.
Park the phone when you finish.

For a first session without an APK, ask it to open Android Settings, read the visible options, and report what it found. Keep one agent in control of each phone.

For your own app, the Android build testing guide explains APK installation and walking through the screens you changed.

Read, act, check

The useful loop is small: observe the current screen, choose an action from what is visible, then observe again to check the outcome.

The CLI returns a text view of the screen with numbered elements. Your agent can use visible text to target a tap, or use a numbered reference together with its snapshot.

phonebox look
phonebox tap --text "Search apps"
phonebox type "chrome"
phonebox look

This sequence comes from the quickstart; the visible screen on your device decides which action is appropriate. Read the observation format and action targeting rules for details.

When the outcome is unclear, look again before continuing. Do not repeat a write action just because a response was delayed or lost.

Park the phone when done

A running phone costs $0.06 per minute, billed per second with a one-minute minimum per session. Parked phones cost nothing and retain their apps and data. New accounts start with $2 in credit and no card requirement.

phonebox park

The default idle timeout is five minutes and the default maximum session is fifteen minutes. The maximum may shorten to fit the available credit. Explicitly parking the phone avoids paying for the remaining idle time.

You can check phone status and session costs in the console. The pricing documentation and cost guide explain reservations and limits.

Choose your interface

The setup prompt is a straightforward starting point for terminal-based agents. When you integrate Phonebox into a workflow, use the interface that fits the agent you already have.

• CLI — drive a phone from the terminal.
• MCP — give your agent phone tools.
• TypeScript SDK — add phone control to your app.