Controls Expo Go on an Android emulator or USB-connected phone through adb, supporting device listing, target selection, key presses, and screen interactions after user consent.
Enables AI to render and interact with Expo/React Native apps headlessly, including PNG screenshots, taps, typing, scrolling, screen descriptions, and automated monkey testing; optionally drives Expo Go on Android and iOS targets.
Controls Expo Go in a booted iOS Simulator using xcrun simctl and idb, supporting taps, typing, scrolling, UI tree inspection, and allowed key presses on macOS with Xcode.
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
An MCP server that lets an AI see and operate an Expo / React Native app without an emulator, a phone or a browser — real PNG screenshots, taps, typing, scrolling and automatic bug hunting. (Русская версия: README.ru.md)
Metro + Expo Go need a native runtime, so instead:
esbuild bundles your project (App.* or main), swapping react-native for a light shim (src/shim/rn.js) and native-only packages (expo-*, react-native-*) for stubs.
The app runs in Node; React renders into our own tree (react-reconciler).
Layout is computed by Yoga (the flexbox engine React Native uses) and painted by Skia (@napi-rs/canvas) into a PNG.
Limits: an approximation, not real iOS/Android — fonts, shadows, animations and native modules differ; expo-router / react-navigation are not supported; state resets on refresh_screen.
Related MCP server: React Native Debug MCP
refresh_screen — rebuild and show the screen (projectPath, entry, width, height, scale, reload)
click_screen — tap at x,y or by target (testID / text), long, or alert:<button>
type_text — type into a TextInput (target or x,y, submit, append)
scroll_screen — scroll the area under x,y
describe_screen — text outline of elements + layout warnings (overlap, clipped text, off-screen)
auto_test — monkey-test everything: several rounds of reload, fill every input with varied data (empty, invalid, valid, very long, special characters, Cyrillic, CJK, Arabic/Hebrew, emoji), tap every button/switch (scrolling to reach it), dismiss alerts. Reports crashes, console errors, layout problems and dead buttons with screenshots.
The server ships built-in instructions: after any code change the AI must rebuild, check by hand, run auto_test, fix and repeat until a full pass is clean. The MCP prompt verify_changes triggers the same loop.
The server's text is English, and the AI answers in whatever language the user writes in. Rendering falls back to system fonts for Latin, Cyrillic, CJK, Devanagari, Arabic, Hebrew and emoji (install Noto fonts on Linux). Right-to-left layout (I18nManager) is not simulated; RTL text itself renders.
claude mcp add expo -e EXPO_PROJECT=/path/to/app -- npx -y expo-headless-mcpTests: npm test (end-to-end), node test/mcp.js (stdio).
Besides the headless renderer, the AI can drive Expo Go on an Android emulator or a USB phone through adb (Windows, macOS, Linux) and in a booted iOS Simulator (macOS with Xcode only). Physical iPhones are not supported: Apple does not allow remote input without heavy tooling — open Expo Go on the iPhone yourself.
Tools: device_list (read-only), use_target (device / headless), press_key; refresh_screen, click_screen, type_text, scroll_screen, describe_screen then act on the device.
Safety rules built into the server:
Nothing is connected / adb missing → it says so, changes nothing, and recommends that you check the app yourself (or use headless mode).
USB debugging off or unauthorized → it explains how to enable it and warns up front that it will tap, type and scroll on your phone.
Consent first. It asks you (MCP elicitation, or via the AI in chat if your client lacks it) before controlling a physical phone, before installing Expo Go when it is missing, and before replacing an Expo Go whose SDK does not match the project (the matching build comes from Expo's official release list; a downgrade needs an uninstall first, which is stated in the question). If you decline, nothing is done and it will not retry.
It only touches Expo Go: every tap/type/swipe first checks that Expo Go is in the foreground. Allowed keys: back, enter, tab, delete.
auto_test is headless-only (it taps everything). Non-ASCII text cannot be typed through adb — test other languages in headless mode.
Status: the Android path has been used on a real phone by the author, and is covered by automated tests against a fake adb.
Starting the dev server runs npx expo start in your project (it asks before running npm install if dependencies are missing). Override with EXPO_MCP_METRO_CMD; set EXPO_MCP_ADB to use a specific adb.
Needs macOS + Xcode (xcrun simctl) and Facebook idb for taps/typing/UI tree (brew tap facebook/fb && brew install idb-companion && pip3 install fb-idb). Without idb the server only tells you how to install it; it never installs it.
Expo Go for the project's SDK is installed/replaced only after you consent (official build from Expo's release list).
The simulator is disposable, so there is no per-app focus check; there is no hardware Back key on iOS.
Status: the iOS path is covered by tests against fake xcrun/idb only — it has not been run on a real Mac yet.
This server cannot be deployed
Drive and test your React Native and Expo apps end-to-end on iOS and Android. Supports real devices, simulators, and emulators. Snapshots, taps, typing, swipes, screenshots, deep links, reloads, builds and runtime errors.
Open a React Native app in a simulated phone: read the screen, tap, type, set device state.
Drive real devices from your AI Coding tool. Embed a client SDK (Unity, Godot, Flutter, iOS/macOS, Android, React Native, Web) in your app, then capture screenshots, traverse the UI tree, inject taps and key events, and run automated test tasks on the physical device over a secure relay.
Give your AI agent a real cloud Android phone: push APKs, drive it, get screenshots and logs.
