Gemini Google-OAuth MCP
Impossible-Studio-TW/claude-gemini-mcp
Use Gemini no Claude com sua conta Google — sem API key, sem custos
Ver no GitHubServidor MCP que conecta Claude ao Google Gemini através de OAuth, sem necessidade de chave de API ou billing. Reutiliza sua conta Google existente e quota Gemini (Advanced, Workspace ou grátis) diretamente no Claude.
Instalação
npx convez-ai Impossible-Studio-TW/claude-gemini-mcp1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Obter segunda opinião do Gemini dentro do Claude sem sair da interface
- Aproveitar quota Gemini Advanced ou Workspace já contratada
- Processar imagens com Gemini mantendo contexto no Claude
- Trabalhar offline de billing ao usar conta Google pessoal
Veredito

Integração sólida e criativa que resolve um gap real: reutilizar quota Gemini já paga sem duplicar credenciais. Setup com venv automático é conveniente, mas exige Node.js + Python 3.10+ e OAuth initial do gemini-cli. Ideal para quem já usa Gemini e quer loop rápido no Claude.
README
A tradução para o português deste README ainda está na fila.
Gemini Google-OAuth MCP
Use Google Gemini in Claude with your Google account login — no API key, no billing.
中文版說明請見 README.zh.md。
What is this?
A Model Context Protocol (MCP) server that lets Claude talk to Google Gemini through the official gemini CLI's OAuth login, instead of a billed Gemini API key. If you already use Gemini in your browser or via the gemini CLI, this plugin lets Claude tap the same quota — Gemini Advanced, Workspace, or free-tier — with zero extra setup.
Built for people who want a fast Gemini second-opinion loop in Claude without standing up Google Cloud billing.
Features
- Two MCP tools: text-only (
gemini_prompt) and multimodal (gemini_with_image) - Two slash skills:
/geminiand/gemini-with-image - Default model
gemini-3.1-pro-preview; switch per-call togemini-2.5-proorgemini-2.5-flash - Wraps the user's existing OAuth — no API key, no service account
- Self-bootstrapping Python venv on first launch (~30 s once, instant after)
- Workspace-trust bypass baked in so the wrapper works from any cwd
Quick Start (Cowork users)
- Download
claude-gemini-mcp-v0.1.0.pluginfrom Releases. - In Cowork: Plugins → Install → Select file.
- One-time auth in Terminal:
npm install -g @google/gemini-cli gemini # pick "Login with Google", finish OAuth in browser - Restart Cowork (Cmd+Q, reopen).
- Try it:
用 Gemini 問: 一句話介紹量子計算
Quick Start (Claude Desktop / Claude Code users)
For users not on Cowork:
- Clone:
git clone https://github.com/Impossible-Studio-TW/claude-gemini-mcp.git cd claude-gemini-mcp - Install Gemini CLI and complete OAuth once (same as step 3 above).
- Register the server in
~/Library/Application Support/Claude/claude_desktop_config.json:{ "mcpServers": { "gemini": { "command": "bash", "args": ["/absolute/path/to/claude-gemini-mcp/run.sh"] } } } - Restart Claude. The first call auto-creates a venv (~30 s); subsequent calls are instant.
Configuration
The plugin does not manage auth itself — it inherits whatever state the gemini CLI has on your machine. Recommended path is OAuth personal login (the no-API-key path). The wrapper sets GEMINI_CLI_TRUST_WORKSPACE=true so the CLI's trusted-folders prompt never blocks the subprocess; remove that line in server.py if you'd rather trust folders manually.
Prerequisites:
- Python 3.10+ (
brew install python@3.12on macOS) - Node.js for the Gemini CLI install
- A Google account that can use Gemini in the browser (free tier or paid plan)
Tools (MCP API)
gemini_prompt(prompt: str, model: str = "gemini-3.1-pro-preview", timeout: int = 120) -> str
gemini_with_image(prompt: str, image_path: str, model: str = "gemini-3.1-pro-preview", timeout: int = 180) -> str
gemini_prompt — single-turn text question; returns Gemini's text response or an error string starting with [.
gemini_with_image — attaches a local image via the CLI's @filepath syntax. Accepts png / jpg / jpeg / gif / webp. Best for OCR, design feedback, and visual question answering, where Gemini's vision is particularly strong.
Slash skills /gemini and /gemini-with-image are convenience wrappers that route directly to those tools.
Demo
Demo coming soon — see RECORD_DEMO.md for the recording script.
Why this exists
I'm a non-developer who uses Claude as my daily creative ops tool, and I wanted a Gemini second-opinion button without standing up GCP billing for a personal account. Every existing Gemini-MCP project I found assumed you'd pay for the API. This wraps the gemini CLI's existing OAuth so Gemini becomes free real estate inside Claude — same quota I already pay for through Google. Sharing it in case anyone else hit the same wall.
Limitations & known caveats
- One-shot only. Each call is independent. No multi-turn Gemini sessions — use Claude for the conversation and call this for atomic queries.
- No tool / function calling. Wrapper exposes
-p(prompt) mode only, not Gemini's agentic tool use. - Imagen image generation isn't reachable on the OAuth path. Gemini OAuth quota cannot drive Imagen — use a separate API-key plugin or another image MCP for that.
- Quota. Subject to your Google account's Gemini quota. Free-tier daily limits apply.
- First call delay. ~30 s on first run for the venv bootstrap; instant afterwards.
- OAuth refresh. If the CLI prints auth errors, run
geminiinteractively once to refresh the token.
Contributing
See CONTRIBUTING.md. Issues, PRs, and feature suggestions are welcome — this is a personal project shared as-is, maintenance is best-effort.
License
MIT — see LICENSE.
Author
- Built by Icy — Impossible Studio TW / Taipei, Taiwan
- GitHub: @Impossible-Studio-TW
- X / Threads / IG: TBD
Issues, PRs, and "this just saved me $20/mo" notes all welcome.