Use Gemini through your browser, without paying for the Gemini API.

Connect Claude to your Gemini web session.

Gemini UI Bridge opens Gemini in a local browser, sends the prompt, waits for the result, saves the output, and returns it to your AI app.

try the simulator

scroll and try the demo

Honest Overview

Useful for personal workflows, not a production API.

The bridge uses the Gemini website like a careful human would. That makes it helpful for a local assistant setup, but login checks, UI changes, and some media downloads may still need your attention.

Personal assistant7.5/10

Creative workflow8/10

Production backend2/10

API replacement3/10

How It Works

Tap a step and watch where the request goes.

The bridge lets another AI use Gemini through your signed-in browser session, without giving that AI your password, cookies, or a paid API key.

Current step Claude asks the bridge

Claude MCP Browser Gemini Archive

Claude sees tools like gemini_prompt and gemini_create_image. It reads the guide first so it does not waste your Gemini runs.

Interactive Demo

Run a simulated Gemini job and see what happens.

This demo does not call Gemini. It shows the bridge flow before you connect the real browser session.

Your idea

1Claudereads guide

2MCPchooses tool

3Browserpastes prompt

4Geminicreates result

5Archivesaves files

Waiting Press run and watch the flow

You will see which tool gets used and where the output is saved.

What It Looks Like

The browser, saved output gallery, and Claude tools.

These are the main parts of the workflow: Gemini answering, outputs being saved, and Claude seeing the MCP tools.

Gemini UI prompt and response — Gemini answers inside the browser.

Saved-output gallery preview — The output gallery keeps runs easy to review.

Claude MCP tools preview — Claude gets tools instead of guessing what to click.

Tool Console

Pick what you want Gemini to do.

Claude gets focused tools instead of guessing where to click. Press a tool to see when it should be used.

Selected tool gemini_prompt

Normal text chat through Gemini web. This is the best first test after login.

Security

Keep private browser data out of GitHub.

This project controls a real browser profile. That profile is basically a logged-in browser, so it must stay local.

Tracked source code

README, scripts, MCP server, static site, examples.

Ignored .env

Local settings and optional bridge tokens.

Ignored profiles/

Browser cookies and Gemini login session.

Ignored stuff/

Private prompts, screenshots, generated media.

Safe push checklist

npm run security-check
git status --ignored
git push

Setup Path

Install, log in, then connect the MCP server.

After login, add the MCP config to Claude Desktop and ask it: "Call gemini_read_guide, then gemini_status."