Image: Android Developers Blog / Google Google's Android CLI 1.0 Gives AI Coding Agents Direct Access to the Android Toolchain
Google shipped Android CLI 1.0 stable at I/O 2026. The command-line interface gives AI agents like Claude Code, Codex, and Antigravity semantic symbol resolution, Compose preview rendering, and UI test execution — with 70% less token usage than running an agent inside Android Studio.
Google announced Android CLI 1.0 stable at Google I/O on May 19. The release turns the Android toolchain into something AI agents can actually use — without requiring Android Studio to be open.
The core idea: agents like Claude Code, OpenAI’s Codex, and Google’s own Antigravity can now interact with Android Studio’s capabilities through a command-line interface built for machine consumption rather than human interaction. You don’t need Android Studio running. You do need the CLI installed.
What agents can do with Android CLI
The main new capability is the android studio command, which gives agents access to tools that previously only existed inside Android Studio’s IDE:
analyze-file— runs Android Studio’s built-in static analysis inspections on a file and returns errors and warningsfind-declaration— locates the exact definition site of a symbol using semantic resolution, not grepfind-usages— finds all references and declarations of a symbol across the project using the IDE’s semantic understandingrender-compose-preview— renders a Jetpack Compose@Previewand returns the rendered image and UI hierarchyversion-lookup— fetches current version information for specific app dependencies from common repositories
There’s also Journey support: agents can execute natural language-driven UI tests. You describe a user flow in plain language and the agent runs it.
Why this matters for token efficiency
Running an AI agent inside Android Studio means the agent has to work through the IDE’s GUI layer — clicking, reading visual output, processing screenshots. That’s expensive in tokens and slow.
The Android CLI is built to be machine-readable from the start. Google says the interface reduces LLM token usage by more than 70% compared to running an agent inside the IDE, and tasks complete three times faster. For teams building Android apps with agentic tools, this is the difference between feasible and impractical at scale.
Bundled with Antigravity 2.0
Android CLI 1.0 ships bundled with Antigravity 2.0, Google’s desktop agent environment. If you’re already using Antigravity, the Android CLI capabilities come with the update.
For users of Claude Code, Codex, or other agents, Android CLI installs separately:
# macOS
brew install android-cli
# Ubuntu/Debian
apt-get install android-cli
# Windows
winget install Google.AndroidCLI
Installation uses your home directory by default, so no admin privileges are required.
New Android Skills libraries
Alongside the stable release, Google added six new skill libraries for agents working with Android:
- Display Glasses support
- CameraX migration tools
- Perfetto SQL queries for performance tracing
- Adaptive UI patterns
- Testing setup guidance
- Styles API migration
These are available as context packages that agents can load when working on projects that use these subsystems.
The bigger picture
Android CLI is part of a pattern Google has been building since Google I/O 2025: making its development tools work with AI agents rather than assuming the agent has to work around them. The same week, Google shipped Antigravity 2.0, expanded the Gemini API’s Managed Agents, and released Gemini 3.5 Flash with strong coding benchmark scores.
The Android CLI specifically matters because Android development has historically been hard for AI agents. The toolchain is complex, the build system is Java-based, and the feedback loops in Android Studio have relied on visual UI state that’s expensive for agents to read. Making the key diagnostic tools accessible via a fast, token-efficient CLI removes a real barrier for anyone who wants to use current coding agents on Android projects.
Sources: Android Developers Blog, TechCrunch, The Next Web