Gemini CLI Pro

Python-based CLI for coding automation with Gemini, dynamic Flash/Pro routing, and local RAG synchronization.

Requirements

• Node.js 18+
• Python 3.10+
• GOOGLE_API_KEY configured
• Python dependencies: google-generativeai, rich, prompt_toolkit, difflib (Python standard library)

Global Installation (npm)

Direct option:

npm install -g gemini-cli-pro@latest --force

Automatic fallback flow:

npm run install:global

npm Publishing

Before publishing, the script enforces the package name:

• gemini-cli-pro

npm run publish:npm

Run

gemini

Chat commands:

• /auto-model
• /models (opens interactive Gemini model selector)
• /model <exact-name> (locks a model manually by exact ID)
• /skills (lists local and global skills)
• /skill <id> (forces a skill for the next interaction)
• /sync
• /reload-mcps
• /status
• /clear
• /clear-history
• /undo (reverts the last file change made by write_file/edit_file in the current session)
• /help
• /exit

Optional MCPs

The file ~/.cache/gemini-history-chats/config.json controls MCPs (local_rag, github, memory). If any MCP is unavailable, the CLI continues running in API-only mode. The native local RAG MCP uses rag-codebase via mcp-rag-server when installed. On every new session, MCPs are reloaded automatically at startup.

Dynamic Skills

The CLI loads skills from:

• Local: .skills/ in the current project directory
• Global: ~/.cache/gemini-skills/

Each .md/.txt file is treated as a skill, and the file name becomes the skill ID.