What is Google Flow Browser MCP?
Google Flow Browser MCP is an open-source Model Context Protocol (MCP) server developed by hitjcl that connects modern AI coding agents—such as Claude Code, Codex, and Cursor—to Google Flow via a visible, local Chrome or Microsoft Edge browser window. Because Google Flow lacks a stable public API for automated generation, this project provides a controlled, secure bridge that allows agents to manage projects, inspect pages, queue reference media, and trigger generation runs without exposing credentials.
Core Technical Capabilities
- 22 Built-in MCP Tools: Comprehensive toolset covering project tracking, DOM inspection, reference asset uploads, queue monitoring, and media downloads.
- Assisted Browser Authentication: Users log in directly through their normal Google account in a visible browser session; the server never touches, records, or transmits passwords, SMS codes, or 2FA tokens.
- Explicit Confirmation Safeguards: Destructive or credit-consuming actions (such as initiating new video renders) require explicit user approval before execution.
- Durable State & Media Verification: Uses SQLite for local task state persistence and incorporates local FFprobe validation to verify downloaded media integrity before marking workflows complete.
Installation & Quickstart
You can connect Google Flow Browser MCP to Claude Code with a single CLI command on macOS, Linux, or Windows WSL:
claude mcp add --scope user google-flow -- npx -y google-flow-browser-mcp
Prerequisites include Node.js 22.13+, Google Chrome or Edge, FFmpeg/FFprobe for media validation, and an active Google account with Google Flow access.
Google Flow MCP vs Manual Workflows
Compared to manual browser creation, integrating Google Flow with Claude Code streamlines multi-step media production pipelines:
- Automated Asset Pipelines: Coding agents can generate code, prepare image prompts, dispatch them to Google Flow, and download resulting assets straight into the active project folder.
- Session Recovery: The SQLite persistence layer enables automatic progress recovery if a browser disconnects or generation queues stall.
- Total Security Compliance: Local execution model ensures authentication cookies, local databases, and temporary videos never leave your local environment.
Technical Specifications
| Category | MCP Servers / AI Agent Extensions |
|---|---|
| License | MIT License (Open Source) |
| Primary Stack | TypeScript, Node.js, Puppeteer / Playwright, SQLite |
| Compatible Clients | Claude Code, Codex, Cursor, Windsurf, Cline |
| Official Repository | https://github.com/hitjcl/google-flow-mcp |
Frequently Asked Questions
Does Google Flow Browser MCP store my Google account password?
No. The server runs purely locally and opens a visible browser window where you sign in manually. It never asks for, reads, stores, or transmits passwords, verification codes, or 2FA tokens.
Which AI agent clients support this MCP server?
It is designed primarily for Anthropic Claude Code and Codex, but can be configured in any MCP-compliant client including Cursor, Windsurf, and Cline.
Will this MCP automatically consume my generation credits?
No. The safety model mandates explicit user confirmation before any prompt or submission that consumes credits is dispatched to Google Flow.