# gemini-skill [Health: Active]

**Category:** 🎥 Multimedia Process  
**Repository:** https://github.com/WJZ-P/gemini-skill  
**GitHub Stars:** 832  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/gemini-skill

## Description
gemini drawing MCP & skill through browser, can be used in openclaw or any agent that supports MCP. Gemini画图 MCP和sill，支持龙虾或任何agent使用٩(๑>◡<๑)۶

## Claude Desktop Quick Installation
Remote MCP endpoint (confidence: high). Install path detected from listing signals. Add as a URL/SSE server in your client:

```json
"mcpServers": {
  "gemini-skill": {
    "url": "https://api.atlascloud.ai/v1"
  }
}
```

## Documentation

## What gemini-skill does

gemini-skill connects MCP clients to the Gemini web application at `gemini.google.com` through a locally managed Chrome, Edge, or Chromium browser. Its browser automation covers AI image generation, multi-turn text conversations, reference-image uploads, extracting images from conversations, downloading full-size generated images, and managing Gemini sessions. Downloaded images can also be processed to remove the Gemini watermark.

The project includes an additional Atlas Cloud provider. This provider uses an OpenAI-compatible API to list available models, send non-streaming text requests, and consume streaming responses into an aggregated result. It is separate from the browser automation path.

## How it works

The MCP protocol layer runs over stdio and coordinates calls through the browser operation and connection layers. A daemon manages the browser as an independent process. When a tool call needs a browser and the daemon is not running, the project starts it automatically. The browser remains available between calls and is released after the configured idle period, which defaults to 30 minutes.

Communication with the daemon uses HTTP for acquisition and status operations, while browser control uses a WebSocket CDP connection. The browser automation uses `puppeteer-extra-plugin-stealth`. A shared CDP port can also let other browser skills reuse the same browser instance.

## Setup and configuration

Use Node.js 18 or newer and install Chrome, Edge, or Chromium. Clone the project, enter its directory, and run `npm install`. Gemini use requires a Google account already logged in to the selected browser profile. Keep headed mode enabled during initial setup so the login flow is visible.

Configuration can be placed in environment variables or a `.env` file. Process environment values take precedence over `.env.development`, `.env`, and code defaults. Relevant settings include:

- `BROWSER_PATH` for a custom browser executable
- `BROWSER_DEBUG_PORT` for the CDP port, defaulting to `40821`
- `BROWSER_HEADLESS` to select headless operation
- `BROWSER_USER_DATA_DIR` for persistent cookies and login state
- `BROWSER_PROTOCOL_TIMEOUT` for long CDP operations
- `DAEMON_PORT` and `DAEMON_TTL_MS` for daemon behavior
- `OUTPUT_DIR` for generated image files

The optional Atlas provider uses `ATLAS_API_KEY`, `ATLAS_BASE_URL`, `ATLAS_MODEL`, and `ATLAS_REQUEST_TIMEOUT_MS`. OpenClaw browser sessions can be reused by setting `BROWSER_DEBUG_PORT` to `18800`, though the project-managed browser retains its own stealth setup.

## Tools and capabilities

The supported MCP capabilities include:

- Generate images from prompts and download full-size results.
- Hold multi-turn conversations with Gemini.
- Upload reference images for image-based generation.
- Extract conversation images as base64 data or downloaded files.
- Create temporary or new sessions and navigate to historical sessions.
- Read and change the Gemini model selection.
- Read and set thinking depth with `gemini_set_thinking_depth`.
- List Atlas Cloud models with `atlas_list_models`.
- Send regular or streaming Atlas Cloud text requests with `atlas_send_message` and `atlas_stream_message`.

## Limitations and notes

This workflow depends on the Gemini website UI and an authenticated Google browser profile, so browser availability, login state, and UI changes affect operation. Headless mode is configurable but is not recommended for the first login. Atlas Cloud requests require an API key and use a separate provider path from Gemini browser automation. The daemon eventually closes an idle browser according to `DAEMON_TTL_MS`.

_Full upstream README: https://allmcps.com/mcp/gemini-skill/readme_