Short answers about how OmniDebugLink works, what it supports, and what it costs. Anything missing? The quick start and platform guides go deeper.
OmniDebugLink is a relay that connects your AI coding tool to your app running on a real device — over the internet. Your AI sends ordinary MCP tool calls (traverse the UI tree, take a screenshot, tap a button, read logs); the SDK inside your app executes them on the device and returns real data. Calls are synchronous: what the AI sees is what's actually on your device. It also drives automated test runs: the same tool calls let an AI agent execute and verify end-to-end flows on real hardware.
Yes. Any MCP client with an AI agent can run repeatable E2E test flows on real devices: drive the UI, take screenshots, read logs, and check real app state — in natural language, with no test framework to maintain.
No — that's the point. The SDK on the device dials out to the relay, and your AI tool connects to the same relay over HTTPS. Both sides only make outbound connections, so it works across networks, offices, and continents: no shared Wi-Fi, no USB, no port forwarding. See how the remote relay works.
Seven official SDKs: Unity, Android, iOS & macOS, Flutter, React Native, Godot (4.2+), and Web (JavaScript). All expose the same capabilities and integrate with a single line — OmniDebugLink.start("<clientToken>"). See Platforms for per-platform install guides.
No. AI tools never touch device tokens: point your MCP client at the OmniDebugLink MCP endpoint and sign in with your account in the browser (Google or GitHub). After that one sign-in, the tool can drive every device registered to your account. Only the on-device SDK uses a clientToken.
Only what your AI tool explicitly asks for: UI trees, screenshots, logs, and the input events it triggers. All traffic is encrypted in transit, and the device only makes outbound connections — nothing listens on the device. You can revoke any token pair or AI-tool authorization from the console at any time.
They solve different jobs. adb and scrcpy assume a human next to the device on USB or the same network; Appium runs pre-written test suites. OmniDebugLink is built for AI coding tools doing interactive, improvised debugging and testing: synchronous MCP calls, from anywhere, on real hardware. Detailed comparisons: vs adb, vs scrcpy, vs Appium.
Yes. Create one token pair per device — each pair binds exactly one device — and connect as many MCP clients as you like. Every AI tool authorized to your account can drive all devices under it; you pick the target per tool call.
Calls are synchronous — your AI tool blocks until the device replies. When you create a token pair you pick the region closest to your device, so traffic takes the shortest path available. For interactive UI work, the round trip feels close to driving the device directly. See synchronous MCP.
That's your call. Most teams wire it into development and QA builds and leave it out of store releases — the SDK is a single start() call you can gate behind a build flag. It does nothing unless the app is running and connected.
Nothing right now — OmniDebugLink is free during the public beta.