Phonebox

dev.phoneboxv0.1.0更新於 Oct 8, 2026

Cloud Android phones for AI agents: create a phone, read the screen, tap, type, install APKs, park.

已驗證Streamable HTTP可網頁執行Developer ToolsBrowser Automation

概覽

AI 產生的概覽

為 AI 助理提供託管的雲端 Android 手機，可讀取畫面、點按、輸入、安裝 APK，並在閒置時停放。

功能
Phonebox 是託管的 Android MCP 伺服器，為代理提供一台真實的雲端手機。工具包含 create_phone、observe（以編號文字清單加截圖讀取畫面）、act（依序執行最多 20 個動作，例如點按、輸入、滑動、捲動、開啟應用程式、等待元素）、install_app、APK 上傳輔助工具、供真人接手的 live_view_url，以及 start_phone、park_phone、list_phones、get_phone、list_apps、reboot_phone 等生命週期管理工具。不需要 ADB、Android Studio 或模擬器。
適用情境
當代理需要操作真實 Android 裝置時使用：測試剛建置好的 APK、探索應用程式並以精確的畫面文字回報問題，或從 CI、沙箱與遠端開發機自動化行動流程。僅支援 Android，不支援 iOS，也沒有電話號碼、簡訊或通話功能。
執行需求
遠端 Streamable HTTP 端點 Authorization 標頭帶入 Bearer 形式的 Phonebox API 金鑰（pbx_...）。不需要本機執行環境、套件或 OAuth。金鑰可在主控台建立，或讓代理從終端機完成註冊取得；提供含額度的免費方案。
安裝前請注意
Authorization 標頭帶有機密 API 金鑰，請勿寫入共用設定或貼到對話中。依每手機分鐘 0.06 美元計費，按秒計費、最低一分鐘，停放的手機免費但執行中的會產生費用。代理可以安裝應用程式、輸入、點按並改變手機狀態，即時檢視連結也允許他人觀看與操作該手機。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 Phonebox，將其新增到工作區。
  2. 為需要使用其工具的對話啟用該服務。

Web executable，透過 Streamable HTTP。 遠端服務在工作區中設定後即可從網頁執行環境執行。

其他 MCP 客戶端

把它新增到你客戶端的 mcpServers 設定中。

{
  "mcpServers": {
    "phonebox": {
      "type": "http",
      "url": "https://phonebox.dev/mcp"
    }
  }
}

README

Phonebox Android MCP server

A hosted Android MCP server for Claude Code, Codex, Cursor and any MCP client, from Phonebox, the agent-first cloud phone. Your agent gets a real Android phone in the cloud. It can read the screen, tap, type, install your APK, hand off to a human, and park the phone when it's done. You don't need ADB, Android Studio or an emulator.

bash
claude mcp add --transport http phonebox https://phonebox.dev/mcp \  --header "Authorization: Bearer $PHONEBOX_API_KEY"
  • Endpoint: https://phonebox.dev/mcp (Streamable HTTP, bearer API key)
  • Price: $0.06 per phone-minute, billed per second, 1-minute minimum. Parked phones are free and keep their apps and sign-ins.
  • Start free: $2 of credit, no card → phonebox.dev

This repo holds the configs and examples. The server itself is hosted by Phonebox, so there's nothing to run.

Why agent-first

Most cloud phones were built for social-media dashboards or QA test suites and added an API later. Phonebox was built for the agent: your agent signs you up from the terminal, reads the screen as text, gets errors with a next step, and never has a write retried behind its back. Idle phones park themselves for free.

Why hosted instead of an ADB MCP server?

Local ADB MCP serverPhonebox hosted MCP
Where the phone runsAn emulator or USB phone on your machinePhonebox's cloud
SetupAndroid SDK, ADB, a system imageOne URL and an API key
Works from CI, sandboxes and remote dev boxesOnly with nested virtualization or a deviceYes
State between sessionsWhatever your emulator keepsApps, files and sign-ins kept while parked
Human handoffSomeone at the machineA live view link in any browser

Get an API key

The fastest way is to paste this into your coding agent. It signs you up from the terminal, and you never paste a key into the chat:

text
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.

Or create a key in the console and export PHONEBOX_API_KEY=pbx_….

Connect your client

Claude Code — run the command at the top, or commit this repo's .mcp.json. It reads the key from each person's environment:

json
{  "mcpServers": {    "phonebox": {      "type": "http",      "url": "https://phonebox.dev/mcp",      "headers": { "Authorization": "Bearer ${PHONEBOX_API_KEY}" }    }  }}

Cursor: add examples/cursor-mcp.json to ~/.cursor/mcp.json or .cursor/mcp.json.

Codex: add examples/codex-config.toml to ~/.codex/config.toml.

Any other client: point it at https://phonebox.dev/mcp with an Authorization: Bearer pbx_… header. The server is stateless Streamable HTTP and doesn't use OAuth.

Tools

ToolWhat it does
create_phoneCreates a cloud Android phone and waits until it's ready
observeReads the screen as a numbered text list, plus a 720px screenshot
actRuns up to 20 actions in order: tap, type, swipe, scroll, open an app, wait for an element
install_appInstalls from the app library or from your own uploaded APK
create_app_upload / complete_app_uploadUploads your APK with one curl command
live_view_urlA link for a person to watch and control the phone
start_phone / park_phoneResumes a parked phone, or parks it to stop billing
list_phones / get_phone / list_apps / reboot_phoneFind phones, wait for them, list apps, restart a stuck phone

What the agent reads from observe:

text
app com.android.launcher3screen 1080x2400[1] EditText "Search apps" #search clickable editable (540,250)[2] TextView "Chrome" #icon clickable (200,1900)[3] TextView "Settings" #icon clickable (500,1900)

Examples

What it isn't

Android only (no iOS), and there are no phone numbers, SMS or calls. Phonebox gives your agent the device, not telephony.

Learn more

來源：README.md，提交 742ab26

工具

0
工具後設資料尚未被收錄。

版本歷史

1
  1. v0.1.0最新Oct 8, 2026