M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
G

Gemini Skill

Community
Control Gemini web via CDP for AI image generation, chat, image extraction, and more.
How to install ↓ Compare GitHub source repository ↗
Category
Browser #27 of 32
Stars
★ 833 Very popular
Transport
stdio (local process)
Runtime
Node.js 18+
Credentials
No credential needed
License
MIT
Last commit
Tools
17
41FMRS · D

This project offers comprehensive control over Gemini web via browser automation, feature-rich but requiring environment setup and login state, suitable for technical users.

Strongest · Documentation 9/20 Weakest · Reliability 7/20

Reliability
7/20
Security and permissions
8/20
Maintenance
9/20
Documentation
9/20
Setup experience
8/20
Why each score
Reliability 7/20
Source code not executed in this review, but project structure is clean with clear separation of MCP, operator layer, and daemon. README claims full MCP tool registration and includes error handling and timeouts. Deductions: no tests or CI workflows, so no verifiable evidence of key paths working.
Security and permissions 8/20
No malware or credential leakage detected. Communication over localhost, browser user data dir is configurable, daemon has idle timeout. Deductions: no confirmation for any destructive operations, remote debugging port exposed to local processes, and boundary of network access not fully documented.
Maintenance 9/20
Repository is active with recent releases (v1.1.0/v1.1.1), 827 stars, and 9 open issues. MIT license present. Deductions: no clear security response channel or contribution guidelines, and no maintained changelog.
Documentation 9/20
Comprehensive README covering features, architecture, setup, tool list, daemon lifecycle, and notes. Missing troubleshooting guide and cost/limit documentation. Configuration priority explained but no detailed env variable examples.
Setup experience 8/20
Clear installation steps with MCP client configuration example and multiple usage modes. Requires Node.js, a browser, and manual Google login. No automated setup scripts, but steps are simple and documented.

Static review · not runListed 2026-08-07

Read the FMRS scoring method →

Fit and risk

What it can accessWrites / deletes local filesUses the networkControls a browser

Best for

  • Developers or AI agents needing to automate Gemini web operations.
  • Users wanting to leverage Gemini image generation within the MCP ecosystem.
  • Technical users comfortable with configuration and browser automation.

Not for

  • Non-technical users who need simple setup.
  • Scenarios requiring official API support or large-scale production.
  • Users needing high stability and intolerant of browser automation fragility.

Required permissions

  • Requires read/write access to browser user data directory (including login state).
  • Requires local file system access to save generated images.
  • Requires network access to connect to Gemini web and Atlas Cloud API (if used).

Risks and side effects

  • Depends on browser automation; may break if Gemini UI changes.
  • Requires Google account login; account security risk.
  • Browser automation may violate website terms of service.
  • Image generation can be slow and may time out.

Setup

Before you start

Runtime：Node.js 18+

ATLAS_API_KEY optionalsecret Atlas Cloud API key, obtained from the atlascloud.ai console; recommended in .env.development
Other optional settings (11)
BROWSER_PATH optional Browser executable path; auto-detects Chrome/Edge/Chromium if unset
BROWSER_DEBUG_PORT optional CDP debug port, default 40821; set to 18800 to reuse an OpenClaw browser
BROWSER_HEADLESS optional Headless mode, default false; keep false on first run to log in
BROWSER_USER_DATA_DIR optional Browser user data dir storing login state, default ~/.wjz_browser_data
BROWSER_PROTOCOL_TIMEOUT optional CDP protocol timeout in ms, default 60000; increase for image generation
DAEMON_PORT optional Browser Daemon HTTP port, default 40225
DAEMON_TTL_MS optional Daemon idle timeout in ms, default 1800000 (30 minutes)
OUTPUT_DIR optional Output directory for generated images, default ./gemini-image
ATLAS_BASE_URL optional Atlas Cloud OpenAI-compatible endpoint, default https://api.atlascloud.ai/v1
ATLAS_MODEL optional Default Atlas Cloud model, default openai/gpt-4o-mini
ATLAS_REQUEST_TIMEOUT_MS optional Atlas Cloud request timeout in ms, default 60000

Prerequisites: Node.js ≥ 18, Chrome/Edge/Chromium installed, and logged into Google account.

  1. Clone: git clone https://github.com/WJZ-P/gemini-skill.git
  2. cd gemini-skill
  3. npm install
  4. Configure environment variables (optional, e.g., BROWSER_PATH, BROWSER_DEBUG_PORT)
  5. Add server to MCP client config, e.g., {"mcpServers":{"gemini":{"command":"node","args":["<absolute path>/src/mcp-server.js"]}}}
claude_desktop_config.json 
{
  "mcpServers": {
    "gemini": {
      "command": "node",
      "args": [
        "<项目绝对路径>/src/mcp-server.js"
      ]
    }
  }
}

Shown for Claude Desktop. Other clients may use a different file or key (VS Code uses "servers") — the configurator below converts it.

.vscode/mcp.json
{
  "servers": {
    "gemini": {
      "command": "node",
      "args": [
        "<项目绝对路径>/src/mcp-server.js"
      ]
    }
  }
}

Goes in your project's .vscode/mcp.json (VS Code uses a "servers" key).

Terminal
claude mcp add gemini -- node '<项目绝对路径>/src/mcp-server.js'

Run it in a terminal; replace any <…> placeholders with your own values first.

Generate and validate client config

Check that it works

After setup, tools like gemini_generate_image and gemini_check_login should appear in the MCP client's tool list; the browser will open the Gemini page on first run—log in to your Google account manually, then calling gemini_check_login and getting a logged-in result confirms the connection.

Troubleshooting

  1. Common issues:
  2. - Login failure: Ensure browser is non-headless, manually login and save user data directory.
  3. - Port conflict: Ensure CDP port (default 40821) is free or change BROWSER_DEBUG_PORT.
  4. - Timeouts: Increase MCP client timeoutMs to 180000ms or more.
  5. - Browser launch failure: Check BROWSER_PATH or install a supported browser.
  6. - Model switch failure: Confirm using new model names (pro/flash/flash-lite).

Things to try

Once connected, you can ask your AI assistant things like:

  • Draw a shiba inu wearing an astronaut helmet and download the full-size image
  • Upload my reference image and generate a cartoon avatar in the same style
  • Start a temp chat and have Gemini draw a cute cat sticker
  • Create a public share link for the latest generated image

Tools 17

gemini_generate_image writes
Full image generation flow with reference image and HD download, takes 60-120s.
gemini_new_chat writes
Start a new blank conversation.
gemini_temp_chat writes
Enter temporary chat mode.
gemini_navigate_to writes
Navigate to a specific Gemini URL (e.g., historical session).
gemini_switch_model writes
Switch model (pro/flash/flash-lite).
gemini_send_message writes
Send text and wait for reply, takes 10-60s.
gemini_upload_images writes
Upload images to the input box.
gemini_get_images read-only
Get metadata of all images in the session.
Show 9 more tools
gemini_extract_image writes
Extract image base64 and save locally.
gemini_download_full_size_image writes
Download full-size HD image.
gemini_share_latest_image writes
Create a public share link for the image.
gemini_get_all_text_responses read-only
Get all text responses.
gemini_get_latest_text_response read-only
Get the latest text response.
gemini_check_login read-only
Check Google login status.
gemini_probe read-only
Probe page element status.
gemini_reload_page writes
Reload the page.
gemini_browser_info read-only
Get browser connection info.

Use cases

AI image generation: create images from natural language prompts.
Text chat: interact with Gemini in multi-turn conversations.
Image processing: upload reference images for image-to-image generation or extract images from conversations.
Automation workflows: integrate Gemini capabilities into OpenClaw or other MCP clients.

Supported clients

Claude Desktop
CodeBuddy
OpenClaw

Listed from the project's documentation, not tested by this site.

Overview

This is an MCP server and skill that controls the Gemini web interface via browser automation (CDP), enabling AI image generation, multi-turn chat, image upload/extraction, session management, and more. It uses a standalone daemon to manage the browser lifecycle, supports automatic watermark removal, reference-image generation, full-size HD downloads, and includes stealth anti-bot plugins. Works with OpenClaw or any MCP-capable client, and also provides an optional Atlas Cloud Provider for OpenAI-compatible API access.

Similar servers

Compare with other MCP servers

Source revision 333e81835c3e Data synced 2026-10-19 Read the FMRS scoring method