Testa MCP server

Testa

Let your AI agent test your iOS app.

Testa gives Claude Code, Codex, Cursor or any MCP client eyes and hands in the iOS Simulator. It reads the screen as compact text, taps, types and swipes with real touch events, and checks the result. It works with React Native, Expo and SwiftUI apps, and you don't have to add a single testID. When the agent is done, save the run as a flow file and replay it in CI without a model.

Install

Requires macOS with Xcode 26 (iOS 26 simulators).

brew tap valewnrt/testa
brew install testa
testa setup        # installs the Claude Code skill and registers the MCP server

Or build from source: git clone https://github.com/valewnrt/testa && cd testa && ./install.sh.

Connect your agent

testa setup already configures Claude Code. For other agents:

codex mcp add testa -- testa mcp      # Codex

{ "mcpServers": { "testa": { "command": "testa", "args": ["mcp"] } } }

The JSON goes into Cursor's .cursor/mcp.json or any other MCP client config. Then ask your agent something like "Launch the app in the simulator, sign up with a test account and check that the welcome screen appears." More in docs/agents.md.

Try it yourself

Testa is a plain CLI, so you can drive it by hand too:

$ testa boot "iPhone 17 Pro"
$ testa install ./MyApp.app && testa launch com.example.myapp

$ testa ui
19 elements (on screen)
e2 StaticText "ready" #status @201,89
e5 Button "Tap me" #tapButton @102,171
e18 TextField #textInput =type here @201,601
…

$ testa tap "Tap me"
tapped e5 Button Tap me
-- ui changes --
~ e2 StaticText "tap:1" #status @201,89
~ e6 StaticText "count: 1" #tapCount @41,208

$ testa assert "#status" label=tap:1
PASS label tap:1

The first command starts a background daemon (a few seconds, once). After that, each command takes about 60 ms.

What makes it different

• No app changes needed. Testa reads the accessibility tree and falls back to on-device OCR (Apple Vision) for anything that isn't in it: system sheets, canvas-drawn screens, games, WebViews. More →
• Built for token budgets. A screen is one line per element, about 200 tokens instead of about 1,500 for a screenshot, and every action replies with only what changed. Benchmark →
• Record once, replay in CI. testa flow record turns what the agent did into a plain-text flow file. testa flow run replays it with JUnit output and failure artifacts, and there's a GitHub Action. More →
• Real gestures and device control. Tap, long-press, swipe, drag-and-drop, pinch, rotate, multi-touch and hardware buttons, plus push notifications, location, Face ID, dark mode, Dynamic Type and more.
• One local binary, no dependencies. Testa talks to Apple's simulator frameworks directly: no Node, no Python, no WebDriver server. No network, no telemetry. How it works →

Status

Testa is pre-1.0 and is used against real production apps. It relies on private Apple frameworks, so a new Xcode can break it; a weekly job runs the suite against Xcode betas so breakage shows up early. Verified on Xcode 26.4 with iOS 26.4 simulators. iOS Simulator only: no real devices, no Android (why).

Documentation

• Commands: selectors, the observe → act → verify loop, full reference
• Using Testa from an AI agent: MCP setup, tools, token cost
• Testing apps without testIDs
• Flows & CI: flow files, recording, GitHub Action
• How it works · Security model
• How Testa compares to Maestro, Appium, idb and Argent
• Troubleshooting & limitations

Contributing

Bug reports, Xcode-compatibility reports and pull requests are welcome. Start with CONTRIBUTING.md; questions and ideas go to Discussions. Changes are listed in the CHANGELOG.

Acknowledgements

Testa's HID and accessibility work was informed by studying Meta's idb (MIT); no idb code is linked or bundled. Thanks to everyone who has contributed.

License

MIT