🖐 MCP-native Android control

ScreenHand gives AI real hands on Android.

Install one app on any Android phone, get a private link, and let your AI agent — or you — see the screen and tap, type, and swipe on real apps with real logins, from anywhere.

⬇ Download for Android Get your link Read the docs
connect your agent — Claude Code
# Add ScreenHand to any MCP client
claude mcp add --transport http android \
  https://mcp.screenhand.com/u/<your-token>/mcp

Works with Claude, Cursor, and any MCP client.  ·  A real device over the internet — built for reliability, not twitch-speed.

Why this exists

Mobile is where the apps are. Agents can't reach it.

🔌

No APIs for mobile apps

The important stuff — social, banking, delivery, logins — lives behind mobile apps with no API. Agents are locked out.

📵

Emulators aren't the real thing

They lack real accounts and real apps, and trip anti-bot checks. For real work you need a real device.

🎯

Pixel automation is fragile

Screenshot-and-guess taps break on every layout change and can't read what's actually on screen.

How it works

A real phone, controllable from anywhere.

Install ScreenHand

Add the app to any Android device and enable it — guided setup: paste your link, turn on accessibility, done.

Get your private link

Each device gets a secret, isolated URL. Your agent controls only your phone — nobody else's.

Connect & go

Point any MCP-compatible AI agent at the link, or open the live view in your browser and drive it by hand.

Your AI agent / browser ⇄ mcp.screenhand.com (secure tunnel) ⇄ Your Android phone   ·   end-to-end over HTTPS
Features

Built to actually work on real devices.

👆

Real taps & gestures

Genuine touch events through Android's accessibility layer — tap, long-press, swipe, drag, pinch, rotate. Not pixel guessing.

⌨️

Real typing

A purpose-built keyboard sends true per-character input, so search-as-you-type and reactive fields actually fire.

🧩

Sees the screen, structurally

Reads on-screen text and UI elements with their positions, so the agent knows what it's looking at.

🔎

On-device text recognition

Reads text from anything — even canvas, WebView and games — using on-device OCR (Google ML Kit). No third-party OCR cloud.

🛠️

20 MCP tools, one URL

A compact, intent-based tool surface — observe, tap, type, swipe, gesture, navigate, find, batch and more. Full power, nothing bloated.

⏺️

Macros — record once, replay free

Record a flow, replay it over a list of inputs, export to CSV. Deterministic, with no AI cost per run.

🔐

Token-isolated

Every device sits behind its own secret link — invalid links see nothing, and one link can't reach another person's device.

🖥️

Live browser view

Watch the device and click-to-tap from any browser, on the same connection — no extra setup.

MCP tools

One link. A full toolset for your agent.

ScreenHand speaks the Model Context Protocol, so any MCP client gets the whole toolset automatically. 20 tools

observe · see screen tap type swipe gesture · pinch/rotate navigate · open app/URL find · wait/scroll to key clipboard file batch run_macro record_macro

Honest note on speed: ScreenHand drives a real device over the internet on a poll loop. It's built for reliability and realness — perfect for app workflows, data pulls, QA and agent tasks — not for real-time, twitch-speed games.

Use cases

What people build with it.

Automate any mobile app

Do on a phone what no API lets you — drive the apps directly.

Pull mobile-only data

Read and export what only exists inside the app, straight to CSV.

Social & growth ops

Run repeatable flows across accounts. Keep within each platform's terms of service.

App QA & testing

Drive your own app on a real device, remotely and repeatably.

Agent workflows

Give an autonomous agent a phone as a first-class tool.

Remote device control

Operate your own device from anywhere, by hand, in the browser.

One agent, two surfaces

One agent. Your phone and your computer.

ScreenHand also gives AI native control of the desktop — macOS and Windows apps through the OS accessibility layer, with the full MCP toolset. Same idea, bigger screen.

Explore desktop GitHub
Integration

Connect in one line.

Point any MCP client at your link and the whole toolset loads automatically.

terminal
claude mcp add --transport http android https://mcp.screenhand.com/u/<token>/mcp
claude_desktop_config.json
{
  "mcpServers": {
    "android": {
      "type": "http",
      "url": "https://mcp.screenhand.com/u/<token>/mcp"
    }
  }
}
terminal
npx -y mcp-remote https://mcp.screenhand.com/u/<token>/mcp
Security & data

Token-isolated — and clear about where your screen goes.

We tell you exactly what's true, and exactly what leaves the device. No "your data stays private" theatre.

Token isolation
Each device sits behind its own secret link. Invalid links see nothing, and one link can't reach another device.
Encrypted in transit
All connections use HTTPS/TLS end-to-end.
You're in control
You explicitly enable ScreenHand and consent before it turns on, and you can stop the worker or disable accessibility at any time.
On-device text recognition
OCR runs on the phone (Google ML Kit) — no third-party OCR cloud.
Where your screen goes (read this)
ScreenHand relays your screen contents over the internet to whatever AI agent / provider you connect it to. It is not local-only, and it is not private from the provider you choose — those screen contents and OCR results reach that provider so it can decide actions. Only run it on a device, and view content, you're comfortable sharing with the agent you connect. Privacy · Terms.
Get started

Get your link and connect in minutes.

Free during beta. Request your private link and we'll get you set up.

  1. Request your link

    Drop your email — we mint a private, device-isolated link for you during the beta.

  2. Install the app & paste your link

    Add ScreenHand to your Android device (direct APK) and follow the guided setup — accessibility, keyboard, battery.

  3. Connect your agent

    Use the one-line snippet for Claude, Cursor or any MCP client — or open the live browser view and drive by hand.

Request your beta link

Tell us a bit about what you'd automate and we'll send your link.

We'll email your private link within 24 hours. Or email singhaldeoli106@gmail.com directly. Two builds: ScreenHand Remote (human-driven, app stores) and ScreenHand Agent (autonomous, direct APK).

Free during beta

Get your link and start now — no card. Pricing comes later, once reliability is hardened on real devices.

Get your link
FAQ

Straight answers.

Does it need root?

No. It uses Android's standard Accessibility Service — no root, and no PC needed after setup.

Which devices?

Any Android phone or tablet on Android 9 or newer.

Is it safe — can it control other phones?

It controls only the device you install it on, behind a secret link. You enable it and can disable it anytime. It's for devices you own or are authorized to use.

Real device or emulator?

A real device — that's the whole point. Real apps, real logins, no anti-bot tripwires from emulation.

Why does it need the accessibility permission?

That's how it reads the screen and performs taps. We disclose this and ask for consent before enabling. Your screen contents are relayed to the AI provider you connect — see Privacy. We never sell your data.

Which AI agents work?

Any MCP-compatible client — Claude Code, Claude Desktop, Cursor and others.

Is it on the Play Store?

The human-driven "Remote" build targets the Play and Galaxy stores; the full autonomous "Agent" build is a direct APK download.

Does running macros cost AI tokens?

No — recorded macros replay deterministically on the device with no AI cost per run.

Put a real Android phone in your agent's hands.

Free during beta. Request your private link and connect any MCP client in one line.

⬇ Download for Android Get your link