Gemini CLI

Gemini CLI is Google’s open-source terminal agent for coding and other command-line tasks. It uses a reason-and-act loop with built-in file and shell tools and can connect to local or remote Model Context Protocol servers. Official documentation presents it as useful for fixing bugs, creating features, improving tests, generating content, researching, and managing tasks. It can run locally after npm installation or directly in Google Cloud Shell, and its canonical repository is licensed under Apache License 2.0.

Access can come from Gemini Code Assist for individuals, Standard, or Enterprise, with quotas shared between Gemini CLI and Code Assist agent mode. A personal Google account provides a free quota, while a Gemini API key supports pay-as-you-go usage. Google-account quotas and available models vary over time and by authentication method; current limits belong on the live Gemini CLI quota and pricing pages. The Google developer page labels Gemini Code Assist agent mode in VS Code as Preview; that status applies to the IDE agent-mode integration, not to the existence of the CLI itself.

Top features

• Interactive terminal agent for repository questions, file edits, shell commands, debugging, and task execution.
• Built-in tools combined in a reason-and-act loop for multi-step work.
• Local and remote MCP server support for extending the agent with approved services and data.
• Project context and commands for inspecting tools, memory, statistics, and MCP connections.
• npm installation through the official @google/gemini-cli package, with an npx path that does not require global installation.
• Availability in Google Cloud Shell without separate local setup.
• Free-quota access with a personal Google account and pay-as-you-go access through a Gemini API key.
• Apache-2.0 source code and public development repository.

Use cases

• Explore an unfamiliar repository and trace likely entry points or bug locations from the terminal.
• Implement a bounded feature, create tests, and react to actual build or test output.
• Use an agent in Cloud Shell, SSH, or another terminal-centered development environment.
• Connect a documented internal service through MCP and use it as context for a coding task.
• Automate content, research, or maintenance tasks that require both language-model reasoning and local command-line tools.