How to Let Claude Code Test Your Mobile App

iOS Simulator MCP is the smallest setup on this list if you only ship to iOS. The agent gets tools to read the accessibility tree (ui_describe_all), tap, type, swipe, take screenshots, record video, install and launch apps and open deep links. Deep links are the underrated part: the agent can jump straight to the screen it changed instead of tapping through the whole app.

It needs macOS, Xcode with simulators, Node 20+ and Facebook's IDB. Install IDB, then add the server:

brew install facebook/fb/idb
claude mcp add ios-simulator npx ios-simulator-mcp

The tradeoff: it is iOS only, it drives the Simulator and not a physical phone, and IDB's current brew formula wants an Apple Silicon Mac on a recent macOS with a full Xcode install, not just the command line tools.

Maestro is an open source UI testing framework for iOS, Android and web, and its MCP server ships inside the CLI. That matters because the agent does two jobs with one tool. While it builds a feature it can inspect the screen, tap, scroll and assert. When you ask for an end to end test, it writes a plain YAML flow, runs it until it passes, and the flow stays in your repo for CI.

Maestro needs Java 17 or newer. Install the CLI and register the MCP server:

curl -fsSL "https://get.maestro.mobile.dev" | bash
claude mcp add maestro -- maestro mcp

The tradeoff: the CLI is free under Apache 2.0, but parallel runs at scale push you toward Maestro Cloud, which is paid. For a solo app the local CLI is plenty.

agent-device from Callstack covers the most ground: iOS, Android, HarmonyOS, TV platforms, web, macOS and Linux, on simulators, emulators and physical devices. Instead of reasoning over screenshots, the agent reads compact accessibility snapshots with refs like @e2, acts on them, and gets a diff of what changed. That keeps token use down on long flows. It also coordinates device access when you run several agents in parallel worktrees.

It needs Node 22.12 or newer (24 for web automation):

npm install -g agent-device@latest
agent-device doctor

For MCP, point your client at the command agent-device with args ["mcp"]. The project also suggests the plain CLI is often enough, since Claude Code can run shell commands directly. The tradeoff is that the breadth means more setup to read through, and it does not replace a test suite on its own, though it can export a run as Maestro YAML.

ARTEMIS is Google's open source Android automation agent. You describe a task or a test in plain English and it drives a real phone or emulator step by step, across apps, with Logcat output and screenshots. It ships an MCP server, a web console, a CLI and a Python SDK, and reports 99%+ on the AndroidWorld benchmark.

Install from the repo, then wire the MCP server into your agents:

git clone https://github.com/google/artemis.git && cd artemis
./start.sh
uv run artemis mcp --install all

The tradeoffs: it is Android only, it runs its own model loop so you pay for the model API you connect on top of Claude Code, and the first task installs an accessibility helper on the device (you can remove it with uv run artemis helper uninstall). It shines for exploratory testing more than for quick checks.

If your "mobile app" is a PWA or a responsive web app, you may not need a simulator at all. Chrome DevTools MCP is Google's official server that lets the agent control a live Chrome: screenshots, console, network requests, performance traces and Puppeteer style automation. The agent can resize the viewport, load the page, read console errors and check that a fix actually removed them.

claude mcp add chrome-devtools --scope user npx chrome-devtools-mcp@latest

There is also a Claude Code plugin that bundles skills (/plugin marketplace add ChromeDevTools/chrome-devtools-mcp). The tradeoffs: it tests Chrome, not Safari on iOS, and usage statistics are on by default unless you add --no-usage-statistics to the args.