โšก Works inside Claude Code, Cursor & any MCP client

Your AI writes the code.
watchr proves it works.

watchr turns your AI agent into a QA engineer that drives your iOS simulators, Android emulators, real devices and web apps โ€” tapping, typing, auditing and catching bugs before your users do. One tool. Every platform.

No credit card ยท iOS ยท Android ยท Web ยท Set up in 2 minutes
claude code โ€” watchr
You โ–ธ QA the signup flow on iOS and find any bugs
watchr โ–ธ launch_app โ†’ observe โ†’ perform(tap "Sign up")
watchr โ–ธ type email ยท tap Continue ยท accessibility auditโ€ฆ
โœ“ 3 screens driven ยท 1 crash caught on empty password
โœ“ 2 a11y issues ยท video saved โ†’ ~/.watchr/videos/signup.mp4
You โ–ธ now run it across 3 Android devices in parallel
watchr โ–ธ parallel_run ยท emulator-5554, 5556, Pixel-7 โ€ฆ
๐Ÿ“ฑ iOS Simulator๐Ÿค– Android๐ŸŒ Web / Playwright๐Ÿ”Œ USB devicesโš™๏ธ MCP
The problem

Shipping mobile & web is fast.
Knowing it actually works isn't.

Your AI agent can build features in minutes โ€” then you spend hours tapping through them by hand. Here's what watchr kills.

๐Ÿ˜ฉ

Manual QA eats your day

Before every release you re-tap the same flows by hand across iPhone, Android and the web. Slow, boring, and you still miss bugs.

โœ“ watchr drives every flow automatically, in seconds
๐Ÿงฑ

Device automation is a maintenance nightmare

Hand-rolling adb, xcrun, idb and Playwright scripts breaks on every Xcode/OS update. You maintain plumbing instead of shipping.

โœ“ watchr wraps it all & keeps it working for you
๐Ÿงฉ

Three platforms, three toolchains

iOS, Android and web each need their own setup, drivers and glue code. Cross-platform QA is fragmented and painful.

โœ“ One unified tool surface for all three
๐Ÿ•ณ๏ธ

Your AI agent is flying blind

LLMs can write the code but can't see the running app โ€” so they can't tell if the button actually works or the screen even loads.

โœ“ watchr gives your agent eyes, hands & a feedback loop
๐ŸŽฒ

Flaky, brittle test scripts

Pixel-coordinate taps and missing waits make tests fail randomly. Nobody trusts the suite, so nobody runs it.

โœ“ Accessibility-aware taps + auto-stabilization
๐Ÿ“ญ

No real QA evidence

"Looks fine on my machine." No screenshots, no video, no audit trail of accessibility, performance or crashes.

โœ“ Video, screenshots, audits & shareable reports
What you get

A full QA team, in one MCP server

Everything your agent needs to drive, observe and audit real apps.

๐ŸŽฏ

Drive any app

tap, type, swipe, scroll, launch โ€” on simulators, emulators, real USB devices and headed browsers.

๐Ÿง 

AI test suites

Describe a flow in plain English; watchr generates, runs and reports a full regression suite.

๐Ÿ“น

Screen recording

Capture video of every run as human-reviewable evidence โ€” iOS, Android & web.

โ™ฟ

Audits built in

Accessibility, performance, security & SEO audits on every screen โ€” automatically.

โš™๏ธ

Parallel & multi-device

Run one flow across many devices at once, or many flows in parallel. One round trip.

๐ŸŽญ

Personas

Test as different users โ€” locales, viewports, timeouts โ€” without rewiring anything.

๐Ÿฉบ

Self-healing setup

watchr --doctor checks your machine and installs what's missing. Zero setup hell.

๐Ÿ’ฅ

Crash & monkey testing

Catches crashes, ANRs and console errors; stress-test screens with chaos input.

๐Ÿ”Œ

Just an MCP server

Plugs into Claude Code, Cursor, Windsurf โ€” any MCP client. No SDK, no rewrite.

How it works

From install to first bug in 2 minutes

STEP 01

Add watchr

One command: claude mcp add watchr -- uvx watchr-mcp. watchr auto-detects your simulator, device or browser.

STEP 02

Just ask

Tell your agent "test the checkout on Android and find bugs." watchr does the tapping, waiting and observing.

STEP 03

Get answers

Screenshots, video, audit results and a report โ€” so you know it works before you ship.

Before & after

Stop babysitting your build

๐Ÿ˜ฎโ€๐Ÿ’จ Without watchr

  • โŒ Hours tapping through flows by hand every release
  • โŒ Brittle adb/idb/Playwright scripts that rot
  • โŒ Separate setup for iOS, Android, web
  • โŒ Your AI guesses whether the UI even works
  • โŒ Bugs found by users, not you

โœ… With watchr

  • โœ… Flows driven & audited in seconds, on demand
  • โœ… One maintained tool surface, always current
  • โœ… One command across every platform
  • โœ… Your agent sees the running app and reacts
  • โœ… Bugs caught before they ship, with evidence
How watchr compares

We're not the only AI QA agent.
We're the one that doesn't make you choose.

Most tools pick a lane โ€” web only, mobile only, or a cloud dashboard that doesn't live in your agent. watchr is the local, in-agent server that covers physical iPhone, Android and web โ€” with audits built in.

Capability Web AI agents
Playwright MCP, Stagehand
Cloud QA platforms
Sofy, Mobot
Other mobile MCPs
Maestro, mobile-mcp
watchr
Drives the web (real browser)โœ“โœ“partialโœ“
Drives Androidโ€”โœ“โœ“โœ“
iOS Simulatorโ€”โœ“โœ“โœ“
Physical iPhone over USB (your device)โ€”vendor cloudsomeโœ“
Runs locally on your machineโœ“โ€”โœ“โœ“
Inside your own AI agent (MCP)โœ“โ€”โœ“โœ“
Built-in audits (a11y ยท perf ยท security ยท SEO)โ€”someโ€”โœ“

Comparison based on each tool's public docs / repositories, verified Jun 2026. Categories: web-only MCP servers (Playwright MCP, Stagehand) drive browsers only; cloud QA platforms (Sofy, Mobot) run on the vendor's servers as dashboards, not inside your coding agent; other local mobile MCP servers (Maestro โ€” iOS sim + Android + Chromium; mobile-mcp โ€” native iOS/Android incl. physical) don't bundle a full web backend + audit suite. Fast-moving category โ€” competitors may close gaps.

vs web AI agents

Past the browser

Playwright MCP & Stagehand stop at the web. watchr drives iOS and Android too โ€” so your agent tests the actual app, not just the site.

vs cloud QA platforms

In your agent, on your machine

Sofy & Mobot are dashboards on the vendor's cloud. watchr runs locally inside your own AI agent โ€” your code and devices never leave your laptop.

vs other mobile MCPs

The whole stack, bundled

Others cover slices. watchr is the one we found that bundles physical iPhone + Android + web + audits in a single local tool.

Pricing

Free to start. Pro when you're serious.

Drive a single device free, forever. Unlock the full QA engine with Pro โ€” 7-day free trial, no card.

Free

$0
core driving on one device
  • observe ยท perform ยท run_steps
  • Single iOS / Android / web target
  • First-run doctor setup
Install free
7-day free trial

Pro

$19/mo
or $190/yr โ€” 2 months free
  • Everything in Free, plus:
  • AI-generated test suites & reports
  • Video, accessibility / perf / security / SEO audits
  • Multi-device + parallel runs ยท personas
Start 7-day trial โ†’
FAQ

Questions, answered

Do I need to write any test code?

No. You talk to your AI agent in plain English โ€” "test the login and find bugs" โ€” and watchr does the driving, waiting, observing and reporting.

Which platforms does it support?

iOS Simulator, physical iPhones over USB, Android emulators & devices, and web apps via Playwright โ€” all through one consistent tool surface.

Where does it run?

Locally, on your machine, as an MCP server. It drives your own simulators, devices and browser โ€” your code and data never leave your laptop.

What do I need to get started?

An MCP client (Claude Code, Cursor, Windsurfโ€ฆ) and one command. Run watchr --doctor and it'll install any missing pieces for you.

What's free vs Pro?

Driving a single device is free forever. Pro unlocks AI test suites, audits, video, personas, multi-device and parallel runs โ€” with a 7-day free trial, no card required.

Get started

Let your agent ship with confidence.

Add watchr to your AI coding setup in one line. Your 7-day Pro trial starts automatically.

claude mcp add watchr -- uvx watchr-mcp