Phonebox

What the Phonebox MCP server does

The Phonebox MCP server provides an Android device in the cloud for AI-driven interaction. An agent can create a phone, wait for it to become ready, inspect the current display, interact with visible controls, install applications, and preserve the device for later work by parking it.

The service is intended for Android workflows rather than general mobile telephony. It does not provide iOS devices, phone numbers, SMS, or calls. The phone, its installed apps, files, and sign-ins remain available while the device is parked.

Typical uses include exploring an APK after building it, checking an Android app through its visible interface, or letting a person take over through a browser when agent control is not sufficient.

How it works

Phonebox is hosted at https://phonebox.dev/mcp and uses Streamable HTTP with bearer-token authentication. There is no local server process to run, and the README states that the service does not use OAuth. A client sends the Authorization: Bearer ... header with a Phonebox API key.

The agent can use create_phone to provision a device and wait for readiness. observe returns the screen as a numbered text representation along with a 720-pixel screenshot. This gives an agent both element-oriented information and visual context. act can execute up to 20 actions in sequence, including taps, text entry, swipes, scrolling, opening an app, and waiting for an element.

Phones can be paused with park_phone or resumed with start_phone. Parked phones are not billed and retain their apps and sign-ins. The listed price is $0.06 per phone-minute, billed by the second with a one-minute minimum.

Setup and configuration

Create an account and API key through Phonebox, then expose the key as PHONEBOX_API_KEY. For Claude Code, the documented HTTP configuration points to the hosted endpoint and sets the bearer header from that environment variable. Cursor and Codex have example configuration files in the repository, while other MCP clients can use the same URL and authorization header.

A client configuration does not require Android Studio, the Android SDK, ADB, a system image, or an emulator. For a private APK, the upload flow creates an app upload and completes the transfer with a curl command. The resulting APK can then be installed on a cloud phone.

Tools and capabilities

The Phonebox MCP server exposes tools for the main device lifecycle and interaction tasks:

• create_phone provisions a cloud Android phone and waits for it to be ready.
• observe reports visible screen elements as numbered text and includes a screenshot.
• act performs ordered tap, type, swipe, scroll, app-opening, and element-wait actions.
• install_app installs an app from the Phonebox library or an uploaded APK.
• create_app_upload and complete_app_upload support transferring an APK.
• live_view_url produces a link for a person to watch and control the device.
• start_phone and park_phone resume or pause a phone.
• list_phones, get_phone, list_apps, and reboot_phone help locate devices, inspect installed apps, wait for phones, or restart a stuck device.

Limitations and notes

The service is Android-only. It supplies device access, not carrier services or mobile identity. Billing applies while a phone is active; parking is the mechanism described for stopping charges while retaining state. Because the server is hosted, local emulator and USB-device workflows are outside its stated setup model.