If you're like me, you pay for one AI subscription and do most of your coding in another tool. I use Claude Code all day, and I also have Google AI Pro. Gemini 4 Argon, Deep Research, Flow, NotebookLM and Jules were all included, and I barely touched them, because switching tabs breaks the flow of work.
So I wrote an open-source MCP server, gemini-pro-mcp, that lets Claude (or Codex, or the Gemini CLI) call the plan directly. Here's what it does and how to set it up in a couple of minutes.
What is MCP, quickly
The Model Context Protocol is a standard way to give an AI agent tools. An MCP server exposes functions like gemini_ask(prompt), and a client such as Claude Code decides when to call them. Install a server once and every MCP client can use it.
What you get
Gemini on your plan, with no API bill. gemini_ask, gemini_research and gemini_code run Gemini through Google's Antigravity CLI, which signs in with your Google account. A second opinion on a tricky diff costs nothing extra.
"Ask Gemini to review this function for race conditions."
Claude calls gemini_ask, reads Gemini's answer and compares it with its own.
Research with sources. gemini_research returns an answer with links, which is handy when Claude needs current information.
Delegated coding. gemini_code lets Gemini edit files in a folder you name, in accept-edits mode with a sandboxed terminal. Claude then reviews the diff. It's a cheap way to split work between two models.
Gemini Notebooks (formerly NotebookLM). Create a notebook, add PDFs or URLs, ask questions grounded in those sources, and generate Audio Overviews, mind maps, quizzes or slide decks.
Jules. Start a Jules session on a GitHub repo, approve its plan and send messages, through the official API.
Flow videos (experimental). flow_make_video turns "a 2 minute video of a lighthouse keeper's day" into a shot list, generates each clip in Google Flow with your monthly credits and joins them into one file. It drives Flow in a separate Chrome profile you sign in to once, and nothing that spends credits runs without an explicit approval call.
Setup
You need a Google AI Pro (or Ultra) plan, uv and the Antigravity CLI signed in.
uvx gemini-pro-mcp setup
This finds the MCP clients on your machine and registers the server in each one. To see what works and what's missing:
uvx gemini-pro-mcp doctor
Restart Claude Code and the tools show up. Try:
"Use gemini_models to list the Gemini models on my plan."
Costs and safety
- Gemini through the Antigravity CLI uses your subscription, not the paid API.
- Video and image tools that use the Gemini API are billed to your API key. Their descriptions say so, and the client asks before calling them.
- Flow tools spend your monthly Flow credits only after a separate approve step.
- Browser tools are off unless you enable them.
Wrapping up
The server is MIT-licensed, on PyPI and in the official MCP Registry:
https://github.com/Muzaqil555/gemini-pro-mcp
If it saves you a tab switch or an API bill, a star helps other AI Pro users find it. Issues and PRs are welcome: there's a guide for adding a new tool in CONTRIBUTING.md.
Top comments (0)