gemini-cli

Gemini CLI is an open-source AI agent that brings Gemini capabilities directly into your terminal. It provides a lightweight, terminal-first interface to Gemini, exposing tools such as grounding with Google Search, file operations, shell commands, and web fetching. Through MCP (Model Context Protocol) support, Gemini CLI can be extended with custom integrations and workflows, enabling automated tasks and scripted interactions from the command line. Use it to query codebases, automate repetitive tasks, fetch real-time information, and orchestrate multi-step workflows from scripts or interactive sessions.

To use Gemini CLI, install or run it via npx, npm, or your preferred package manager, then invoke gemini with optional flags. You can run in interactive mode for conversational prompts or in non-interactive mode to obtain structured outputs suitable for scripting. Typical usage includes starting a session, selecting a model, and issuing prompts that Gemini will answer while grounding responses with real-time information when needed. The MCP support allows you to connect Gemini CLI with other services or custom capabilities, expanding its reach beyond the built-in features.

Prerequisites:

• Node.js (14.x or newer) and npm/yarn installed
• Basic familiarity with the command line

Installation steps:

1. Install or verify Node.js and npm:

   • macOS/Linux: follow nodejs.org installation instructions
   • Windows: install via Node.js installer

2. Install Gemini CLI using npx (no permanent install required): npm run -s if you need an npm script; otherwise you can run directly:

   npx -y @google/gemini-cli

3. (Optional) Install globally for easier access: npm install -g @google/gemini-cli

4. Configure authentication (environment variables):

   • For API key: export GEMINI_API_KEY=YOUR_API_KEY
   • For Vertex/Google integration: export GOOGLE_API_KEY=YOUR_API_KEY and export GOOGLE_GENAI_USE_VERTEXAI=true
   • Set your Google Cloud project: export GOOGLE_CLOUD_PROJECT=YOUR_PROJECT_ID

5. Run Gemini CLI: gemini

Note: If using npx, you can run commands directly without permanent installation.

Tips and common issues:

• If you encounter authentication errors, verify that the correct environment variables are exported and accessible in your shell.
• For MCP integrations, ensure the target server name (gemini-cli) matches in your MCP config and that the command is visible in your runtime environment.
• When using Vertex AI, ensure your Google Cloud project has the appropriate permissions and API access enabled.
• The Gemini CLI supports a wide range of prompts; for scripting, use --output-format json or --output-format stream-json to get structured results suitable for parsing.
• If using npx in constrained environments, consider a local npm cache or a temporary installation to avoid network-related delays on every invocation.