Google opens Android Studio and CLI to Claude Code and Codex

Google on October 28 made Bring Your Own Agent stable in Android Studio Rabbit 2, opening the IDE to more than 40 coding agents that speak the open Agent Client Protocol (ACP). A team that already works in Claude Agent or Codex can keep using it and still get what Android Studio knows about the project: its structure and build, symbol resolution, Compose Preview, the Emulator and Logcat.

Agents install from a marketplace under Settings > Tools > AI > Agents. You can add several, switch between them as you work, and get new agent features as soon as their makers ship them, says Google’s Android Dev Summit tools roundup. Gemini comes in the same way: logging into Google Antigravity through Bring Your Own Agent gives access to the latest Gemini models as they launch. In January, when Android Studio added local models through LM Studio or Ollama, Google called Gemini the default recommendation, tuned specifically for Android development. Matthew McCullough, vice president of product for Android, has emphasized letting developers choose their own models and coding agents, according to The News Stack.

Plugins for Claude Code and Codex

Outside the IDE, Android CLI now ships as an official agent plugin for Claude and Codex. Google launched the CLI earlier this year as a tool-first interface for agents. The plugin lets an agent create Android projects, manage emulators and virtual devices, inspect live UI layouts, run Journeys, browse and install official Android skills, and search the Android Knowledge Base. Journeys are natural-language app tests: per Android’s documentation, the instructions are turned into actions, and assertions are checked against what the AI sees on the device. Google lists three ways to install the plugin:

• Inside a Claude Code session: /plugin marketplace add android/android-cli-plugin, then /plugin install android-cli@android-cli-plugin
• Claude from the terminal: claude plugin marketplace add android/android-cli-plugin, then claude plugin install android-cli@android-cli-plugin
• Codex from the terminal: codex plugin marketplace add android/android-cli-plugin, then codex plugin add android-cli@android-cli-plugin

The CLI also picked up commands for agents and CI jobs that run with nobody watching the screen. android device remote provisions a real physical device from a secure Google device lab, and the app is then installed and tested the same way as on a local device over adb. android device hub shows local and remote devices and emulators without opening an IDE. android preview renders Compose Previews without an emulator or the IDE, so an agent can check its UI changes across screen sizes as it iterates. And android profiler now runs on Android Performance Analyzer, the profiler behind Android Studio, to capture system traces from the command line or in CI. Paired with the new android-profiler skill, an agent can capture a trace, analyze it, propose a fix and measure its impact. The News Stack adds that agents also get access to production crash reports.

Reviewing what the agent changed

Android Studio now has Change Review, which produces a report of potential issues in pending agent changes or in the Git changes window. You can start it with /review from the agent tool window, ask for an AI-guided walkthrough of a long change, and have the agent fix what the review found. The January update gave Agent Mode a changes drawer, where an agent’s edits can be kept or reverted individually or all together.

To help pick a model, Google publishes Android Bench 2.0, a leaderboard of models and agent harnesses on long-running Android tasks, such as migrating an app’s UI or building an app from mocks. Google says these are tasks that take an Android engineer days. Each solution is graded on whether it actually works, plus cost and time.

Google’s figures and caveats

In Google’s internal testing, agents that used android preview completed UI tasks 77% faster while using up to 3x fewer tokens. The News Stack notes that the company shared little about how it measured those results. The memory-leak figures are still Google’s own too: with Android CLI’s LeakCanary tool, the most capable models ran the find-and-fix process almost 2x faster and used up to 25% fewer tokens than without Android CLI. Google illustrates that comparison with Claude Opus 5.5.

Not all of it is stable. Google describes LeakCanary as integrated into Android CLI as android profiler leakcanary, which analyzes heap dumps on your computer instead of the device so the app isn’t slowed down. In the same post, Google says the memory leak subcommand is “coming soon” to Android CLI. Fix with AI for memory leaks in the Studio Profiler is in the Canary build, and so is the Migration Assistant, which converts XML layouts, view-binding code and logic to Compose with the built-in agent or any agent added through Bring Your Own Agent.

In the keynote, Google gave Claude Code a list of critical user journeys and one prompt: find and fix memory leaks in this app. Claude ran each journey on its own device, in parallel, followed each reference chain to fix the underlying code, then re-ran every journey to confirm the leaks were gone.