Google"s Gemini CLI is getting "extensions," a new feature that the company says allows you to "connect your favorite tools and personalize your AI-powered command line."
The extensions are hosted in public GitHub repos, a decision that allows anyone to publish an extension without direct approval from Google. There are over 50 available on the extension gallery, including a Stripe extension for integrating its payment API into agentic workflows and a Terraform extension for infrastructure automation.
Another extension connects Chrome DevTools for coding agents. Google also released its own Nanobanana extension that lets you generate and manipulate images with text prompts directly from the terminal.
Once an extension is installed, Gemini can choose when to use these tools based on your prompt, or you can explicitly invoke them through custom commands that extensions can define. To view the list of extensions you have installed, you can simply use the /extension command.
If you"re a developer, you can create a Gemini CLI extension by starting with one of the built-in templates. The process involves defining the extension"s properties in a gemini-extension.json manifest file, which points to a Model Context Protocol (MCP) server that exposes new tools to the AI. This server is essentially a small application that listens for requests from the CLI. There"s also a commands directory where you can add simple TOML files to create custom command shortcuts for complex prompts.
Gemini CLI was released earlier this year, giving developers a direct way to interact with the AI from their terminal. This initial release already had powerful capabilities, allowing you to do things like ask the AI to read local files or execute shell commands to diagnose problems. Logging in with your personal Google account can get you a generous free tier of 60 requests per minute.
In the months following the launch of the CLI tool, Google has brought it to its Firebase Studio cloud IDE, as well as the Zed Editor, where it can handle tasks like code generation, refactoring existing functions, and providing instant explanations for confusing code blocks.