WJZ-P/gemini-skill

What it solves

Gemini Skill enables the automation of the Gemini web interface, allowing users to perform tasks like AI image generation, multi-turn text conversations, and image extraction without manually interacting with the browser. It bridges the gap between an AI agent (via the Model Context Protocol) and the Gemini web UI, effectively turning the web interface into a programmable tool.

How it works

The project uses the Chrome DevTools Protocol (CDP) and Puppeteer (with a stealth plugin to avoid bot detection) to control a browser instance. It employs a "Daemon" architecture where a background process manages the browser's lifecycle, keeping it open for 30 minutes of inactivity before automatically shutting down to save resources.

It is implemented as an MCP (Model Context Protocol) server, meaning it can be plugged into MCP-compatible AI clients (like Claude or CodeBuddy). The AI client sends requests to the MCP server, which then translates those requests into browser actions on gemini.google.com.

Who it’s for

• AI Agent developers who want to give their agents the ability to use Gemini's web features (especially image generation).
• Power users looking to automate repetitive tasks in the Gemini web interface.
• Developers who want to integrate Gemini's web capabilities into other workflows via a standardized protocol.

Highlights

• Full Automation: Supports image generation, image uploading for reference, and high-definition image downloading.
• MCP Integration: Standardized interface allows any MCP-compatible client to control Gemini.
• Smart Lifecycle Management: A background daemon automatically starts the browser on demand and handles lazy destruction after inactivity.
• Anti-Detection: Uses puppeteer-extra-plugin-stealth to bypass automation checks.
• Image Processing: Includes built-in functionality to automatically remove watermarks from downloaded images.
• Atlas Cloud Support: Includes an optional OpenAI-compatible provider for direct API access to various models.