┌─────────────┐
  │ ■  Settings │
  │ ─────────── │
  │ ◉ Wi-Fi     │
  │ ◉ Bluetooth │
  │ ▸ Display   │
  └─────────────┘

  baremobile

AI agents control your phone like you do -- same device, same apps, same screen. Prunes the accessibility tree down to what matters. Clean YAML, zero wasted tokens.

---

What this is

baremobile gives AI agents full control of real mobile devices -- read the screen, tap, type, swipe, launch apps, send SMS, take photos. The screen comes back as a pruned accessibility snapshot with [ref=N] markers; the agent picks a ref and acts on it.

No Appium. No Java server. No Espresso. Zero runtime dependencies. Same patterns as barebrowse -- agents learn one API for both web and mobile.

Android -- full screen control via ADB, plus on-device APIs (SMS, calls, GPS, camera) via Termux. Use it for QA, as a personal AI assistant, or for remote device management.

iOS -- same snapshot() → tap(ref) pattern via WebDriverAgent. Shared prune pipeline, identical YAML output. No Mac, no Xcode. Designed for QA (USB required on Linux).

Platform	Mode	Where it runs	What it does	Requires
Android	Host ADB	Your computer	Screen control -- snapshots, tap/type/swipe, screenshots, app lifecycle	adb + USB or WiFi
Android	Termux ADB	On the phone	Same screen control, no host machine	Termux + wireless debugging
Android	Termux:API	On the phone	Device APIs -- SMS, calls, GPS, camera, clipboard, contacts	Termux + Termux:API app
iOS	WDA	Your computer	Screen control -- snapshots, tap/type/scroll, screenshots	USB + WDA on device

Host ADB is the default. Termux modes run on the device itself -- useful for a phone that acts as its own autonomous agent. Termux ADB and Termux:API combine for screen control plus device APIs, all from the phone.

Quick start

Prerequisites: Node.js >= 22. Android needs adb in PATH (platform-tools). iOS needs Python 3.12 for setup (runtime is pure HTTP).

npm install baremobile

Three flavors: CLI, MCP server, or library import. Pick one.

CLI

npx baremobile open                       # start daemon
npx baremobile launch com.android.settings
npx baremobile snapshot                   # -> .baremobile/screen-*.yml
npx baremobile tap 4                      # tap ref 4
npx baremobile close                      # shut down

Full command set: open, close, status, snapshot, screenshot, tap, tap-xy, tap-grid, type, press, scroll, swipe, long-press, launch, intent, back, home, wait-text, wait-state, grid, logcat.

MCP server

Claude Code:

claude mcp add baremobile -- npx baremobile mcp

Claude Desktop / Cursor -- add to config (claude_desktop_config.json, .cursor/mcp.json):

{
  "mcpServers": {
    "baremobile": {
      "command": "npx",
      "args": ["baremobile", "mcp"]
    }
  }
}

10 tools: snapshot, tap, type, press, scroll, swipe, long_press, launch, screenshot, back.

Library

import { connect } from 'baremobile';

const page = await connect();              // auto-detect device
const snapshot = await page.snapshot();     // pruned YAML with [ref=N] markers

await page.tap(5);                         // tap element
await page.type(3, 'hello');               // type into field
await page.scroll(1, 'down');              // scroll
await page.launch('com.android.chrome');   // open app
await page.back();                         // navigate back

Works with any LLM orchestration library. Ships with an adapter for bareagent.

Full API, snapshot format, interaction patterns, and gotchas: baremobile.context.md.

What the agent sees

- ScrollView [ref=1]
  - Group
    - Text "Settings"
    - Group [ref=2]
      - Text "Search settings"
  - ScrollView [ref=3]
    - List
      - Group [ref=4]
        - Text "Network & internet"
        - Text "Mobile, Wi-Fi, hotspot"
      - Group [ref=5]
        - Text "Connected devices"
        - Text "Bluetooth, pairing"

Compact, token-efficient. Interactive elements get [ref=N] markers. The agent reads the snapshot, picks a ref, acts on it. Bloated accessibility trees get a 4-step pruning pass, 200+ widget classes mapped to semantic roles. Text input quirks, multi-device setups, element state tracking, and vision fallback are handled automatically.

Device setup

The interactive wizard handles everything -- adb install, SDK setup, device connection:

npx baremobile setup   # Android: emulator, USB, WiFi, or Termux

Manual setup (USB):

1. Enable Developer Options -- Settings > About phone > tap "Build number" 7 times
2. Enable USB debugging -- Settings > Developer options > toggle on
3. Connect device via USB, tap "Allow" on the prompt
4. Verify -- adb devices should show your device

Android 10+ required (2019 or newer). For WiFi, Termux, emulator, and iOS setup details, see docs/customer-guide.md.

Tested against

Settings, Messages, Chrome, Gmail, Files, Camera, Calculator, Contacts, Play Store, YouTube -- on physical devices and emulators across API 33-35.

The bare ecosystem

Three vanilla JS modules. Zero dependencies. Same API patterns.

	bareagent	barebrowse	baremobile
Does	Gives agents a think→act loop	Gives agents a real browser	Gives agents a mobile device
How	Goal in → coordinated actions out	URL in → pruned snapshot out	Screen in → pruned snapshot out
Replaces	LangChain, CrewAI, AutoGen	Playwright, Selenium, Puppeteer	Appium, Espresso, UIAutomator2
Interfaces	Library · CLI · subprocess	Library · CLI · MCP	Library · CLI · MCP
Solo or together	Orchestrates both as tools	Works standalone	Works standalone

What you can build:

• Headless automation — scrape sites, fill forms, extract data, monitor pages on a schedule
• QA & testing — automated test suites for web and Android apps without heavyweight frameworks
• Personal AI assistants — chatbots that browse the web or control your phone on your behalf
• Remote device control — manage Android devices over WiFi, including on-device via Termux
• Agentic workflows — multi-step tasks where an AI plans, browses, and acts across web and mobile

Why this exists: Most automation stacks ship 200MB of opinions before you write a line of code. These don't. Install, import, go.

License

MIT