google-labs-mcp
π§ͺ Google Labs & Flow MCP Server (google-labs-mcp)
The first open-source Model Context Protocol (MCP) bridge connecting Antigravity (AGY) and AI coding agents directly to Google Labs and Google Flow.
Harness DeepMind's bleeding-edge creative modelsβImagen 3 (ImageFX), Lyria (MusicFX), and Gemini Omni Flash / Veo (VideoFX)βdirectly from your coding environment.
Above: High-resolution 16:9 cinematic concept art generated autonomously by Antigravity through this MCP server on Google Flow.
β‘ Why This Matters
AI coding assistants are typically confined to writing text files and code snippets. When designing games, websites, or apps, developers are forced to manually switch between browser tabs, generate assets on web playgrounds, download files, and copy them into their project trees.
google-labs-mcp closes that gap. It equips your agent with direct, autonomous programmatic hooks into labs.google and flow.google.com:
π¨ Instant Concept Art & Textures: Request high-resolution imagery via Imagen 3 directly during coding sessions.
π΅ Dynamic Soundtracks & Stems: Prompt Lyria for custom looping background music and combat SFX.
π¬ Cinematic Video & Cutscenes: Orchestrate video generation models for trailers, cutscenes, and animated backgrounds.
π Zero-Friction Authentication: Automatically attaches to your active Chrome browser session (inheriting Google One / Ultra plan privileges) with zero OAuth hassle or API quota surcharges.
π οΈ Tools Exposed to MCP Clients
Tool Name | Parameters | Description |
|
| Generates cinematic AI video via Gemini Omni 1.1 Flash / Veo 2 on Google Flow using Ultra compute. |
|
| Generates uncompressed imagery via Nano Banana Pro / 2 & Imagen 3 on Google Flow / ImageFX and saves it locally. |
|
| Prompts DeepMind's Lyria (MusicFX) to generate full audio tracks and stems. |
|
| Opens Chrome to |
| (none) | Returns active browser connection status, Flow project ID, Ultra subscription status, and active models. |
π Quick Start Guide
1. Prerequisites
Node.js v18+ (Tested on v24)
Google Chrome browser installed
A Google account (works with standard accounts, with full prioritized compute on Ultra / Google One)
2. Installation
git clone https://github.com/ssfdre38/google-labs-mcp.git
cd google-labs-mcp
npm install3. One-Click Session Launch
Run the session launcher:
start_labs_session.batThis launches Chrome with --remote-debugging-port=9222 and a dedicated profile (~/.gemini/labs_chrome_profile). Log into your Google account once in this window, and your session remains permanently preserved!
4. Register in Your MCP Client
For Google Antigravity (~/.gemini/config/mcp_config.json):
{
"mcpServers": {
"google-labs": {
"command": "node",
"args": ["C:\\path\\to\\google-labs-mcp\\index.js"]
}
}
}For Claude Desktop (claude_desktop_config.json):
{
"mcpServers": {
"google-labs": {
"command": "node",
"args": ["C:\\path\\to\\google-labs-mcp\\index.js"]
}
}
}ποΈ Architecture
google-labs-mcp/
βββ index.js # MCP Server entrypoint & CDP Puppeteer controller
βββ start_labs_session.bat # Chrome launcher with remote debugging enabled
βββ package.json # Dependencies (@modelcontextprotocol/sdk, puppeteer-core)
βββ assets/
β βββ showcase.jpg # Autonomous generation proof of concept
βββ README.md # Documentation & integration guidesπ₯ Contributors & Credits
Concept & Architecture: Conceived and built autonomously by Antigravity (Google DeepMind).
Patron & Visionary: Daniel Elliott (@ssfdre38) β Barrer Software.
Special Thanks: Google DeepMind for creating the incredible generative models powering
labs.googleandflow.google.com.
π License
MIT License. Open source and free for the developer community.