Skip to content

How to Use Android CLI Device Streaming for Remote Android Testing

Learn how to use Android CLI device streaming to reserve real Android hardware, deploy apps through ADB, inspect screens, and install Android skills for AI agents.

How to Use Android CLI Device Streaming for Remote Android Testing

On this page

Google has added Android Device Streaming to Android CLI, so developers can reserve real Android hardware from the terminal and connect it to adb without having the phone physically connected. The October 2, 2026 update also expands Android skills for AI agents, making the command-line workflow useful for both manual development and agent-assisted testing.

What Android CLI device streaming changes

Android CLI is Google's command-line interface for Android development. It can create projects, manage software development kits, start emulators, install applications, inspect interfaces, run tests, and connect with Android Studio. The new device-streaming support adds another option: instead of relying only on a local emulator or a device sitting on your desk, you can reserve a physical Android device hosted in Google's data centers and make it available through the same command-line workflow.

The connection uses Android Debug Bridge (ADB), the standard command-line tool for communicating with Android devices, over a secure ADB-over-SSL connection. Once the remote device is connected, commands such as android run, android install, and other ADB-based tools can work with it much like a locally connected device. Google says the streamed device can also be used for deploying builds, collecting logs and traces, and taking screenshots without opening a device window.

Install or update Android CLI before using remote devices

Start by making sure Android CLI is installed and current. Google currently documents Android CLI as a stable 1.0 tool, and the October 2026 release notes show continued updates to its remote-device workflow. If Android CLI is already installed, use the built-in update command.

android update

Then verify that the command is available from your terminal.

android --version

You should get the installed Android CLI version rather than a shell error. Keeping the CLI updated matters here because remote-device functionality has been arriving through recent releases, including authentication, device catalogs, reservations, and automatic connection behavior.

Sign in before requesting a streamed device

Remote Android devices are associated with a Google Cloud project that has access to device streaming. Android CLI provides an authentication command for signing in to Google services, so the first step is to authenticate from the terminal.

android auth login

After authentication, ask Android CLI which Google Cloud projects are available for device streaming.

android device remote projects

The command helps identify a project that can be used for remote devices. This is separate from simply having an Android application project on your computer: the streamed hardware is a Google-hosted physical device, and its usage is associated with the selected Google Cloud project.

Find a physical Android device and API level

Once a suitable project is available, list the remote device models.

android device remote models

The output provides device identifiers in the / format. That distinction is useful when testing compatibility because the device model and Android API level can affect behavior that a generic emulator may not reproduce.

For example, Google's command documentation shows a reservation such as tokay/34. The important part is not to copy that device blindly into your own workflow; select an identifier that actually appears in your current models output and matches the hardware and API level you need to test.

Reserve and connect the remote Android device

With the project and device identifier in hand, create a reservation. Replace the example device with one returned by the previous command.

android device remote create tokay/34 --project=my-project

Current Android CLI behavior connects the newly created reservation to ADB automatically. Google added that behavior in the October 2026 release. If you want to reserve the device without connecting immediately, pass --connect=false and connect it later using the reservation ID.

android device remote create tokay/34 --project=my-project --connect=false

Then connect the reservation explicitly.

android device remote connect 

A successful connection makes the remote physical device available to ADB. That is the key handoff in this workflow: after the reservation is connected, Android development commands do not need a completely separate remote-device interface.

Check that ADB can see the streamed device

Before deploying an application, verify that the device has appeared in the local ADB device list.

adb devices

The exact identifier shown depends on the streamed device and current connection. The useful result is that the remote hardware appears as an available device instead of leaving the list empty. This is also where an existing ADB workflow can fit naturally; developers already familiar with ADB Wi-Fi will recognize the same basic idea of exposing an Android device to command-line tooling, although Google's cloud streaming service uses its own remote reservation and secure connection path.

If the device does not appear, check the reservation status before changing your application or build configuration. Android CLI provides commands for listing, connecting, disconnecting, extending, and removing reservations, so the device lifecycle can be managed independently of the application itself.

Deploy your app to the streamed hardware

Once ADB sees the device, use Android CLI to build, deploy, and launch your application. The simplest route is the android run command from the project directory.

cd my-android-project android run

Android CLI's run command builds, deploys, and launches an Android application on a connected device or emulator. Because the remote physical device is connected through ADB, it can participate in this workflow alongside local devices.

If you already have an APK and only need to install it, use the dedicated install command instead.

android install app/build/outputs/apk/debug/app-debug.apk

This separation is useful when the build has already been produced by another tool or by continuous integration. You can generate the APK elsewhere, connect a streamed device, install the artifact, and then use Android's normal inspection and testing commands against the real hardware.

Capture screens and inspect the app from the terminal

Android CLI includes commands for inspecting a running application's screen and layout. That becomes particularly useful with remote hardware because you do not need to keep a local phone in front of you while an automated workflow is running.

android screen

The CLI can also inspect the layout hierarchy. By default, the layout command focuses on interactive elements, while the --full option requests the complete layout tree.

android layout --full

These capabilities give an agent or automated script something more useful than a simple β€œthe app launched” result. It can inspect what the application rendered, identify accessible UI elements, and use the resulting information as part of a testing or debugging workflow.

Use Android skills to give coding agents Android-specific guidance

The same October update expands Google's Android skills repository to more than 20 skills. An Android skill is a structured SKILL.md instruction set that gives an AI coding agent Android-specific guidance, examples, and development patterns rather than relying entirely on the model's general knowledge.

Google lists skills for tasks including Android Intent security, CameraX migration, Play policy compliance, R8 configuration audits, testing setup, Android profiling, and Wear OS Compose Material 3. This is separate from simply choosing an AI model or coding assistant. The skill provides additional instructions and Android-specific context that the agent can apply while working on the project.

If you use an AI coding workflow such as the one covered in Android Studio BYOA, Android CLI skills can provide another layer of Android-specific instructions around the agent's normal capabilities.

Install a specific Android skill into your project

Initialize the Android CLI environment first so the base Android CLI skill is available to an agent.

android init

Next, list the available skills.

android skills list

When you find a skill that matches your project, install it into the current project directory. Google's October example uses the Wear Compose Material 3 skill.

android skills add wear-compose-m3 --project=.

The --project=. option keeps the installation associated with the current project rather than treating it as a global installation. The skills system also supports targeting individual agents, and Google documents support for agent environments including Gemini, Claude, and Codex.

Keep installed skills synchronized with the latest guidance

Skills are not static documentation. Google can update their instructions as Android APIs, recommended patterns, and development practices change. Android CLI therefore provides commands for updating them.

android skills update --all

You can also update one skill when you only need a particular workflow refreshed.

android skills update wear-compose-m3

If you customize an installed skill, keep a copy under a different directory name before updating it. Google's documentation warns that adding or updating a skill can replace an existing installed version, so local modifications should not be treated as permanent just because they are currently in the skill directory.

Clean up the remote device when testing is finished

A streamed physical device is a reserved resource, so do not leave reservations running after the test is complete. First disconnect the device if it is still attached to ADB.

android device remote disconnect 

Then remove the reservation.

android device remote remove 

Android CLI also supports listing reservations and extending an active reservation when a test takes longer than expected. Remote devices are billed to the Google Cloud project associated with the streaming session, so the reservation lifecycle is not just housekeeping; it is part of managing the cost and availability of a cloud-based test device.

What this workflow is useful for

The new workflow is most useful when a test depends on real hardware or an Android configuration that is inconvenient to keep locally. A local emulator remains useful for fast iteration, while a streamed physical device can provide another test environment without requiring a developer to own every device needed by the project.

It also changes what an AI coding agent can do. Google describes the remote device as available to agents through Android CLI, which means an agent can work through a sequence that includes building the application, deploying it to physical hardware, collecting diagnostic information, and capturing screenshots. That does not make the agent's output automatically correct; it gives the agent a more realistic environment in which its work can be checked.

For a practical setup, start with one remote device and one repeatable test case rather than immediately automating the entire application. Confirm authentication, reservation, ADB visibility, installation, and cleanup in that order. Once those pieces work reliably, the same command sequence can become part of a larger scripted or agent-assisted Android testing workflow.

Copy this post

0 Comments

No comments yet. Be the first to share your thoughts.

Join the conversation

Log in or create a free account to leave a comment. You can edit or delete your own comments any time.