miyakejima/google-flow-mcp

---

Google Flow MCP is a local Model Context Protocol server that lets an AI agent operate Google Flow through a user-owned Chromium session. It can discover the exact options available to each account, generate media with Flow subscription credits, wait for long-running jobs, use the upscale choices Flow actually offers, and save validated files directly into a project.

It does not require a Google generation API key. Credits are consumed through the Flow website just as they are when the user operates Flow manually.

Installing with an agent? Tell it to read AGENTS.md and follow INSTALL.md. The supported Antigravity command is npm run setup:antigravity. Installer agents must not modify source, install Playwright Chromium, generate tool schemas, or use an Antigravity scratch directory.

Why Google Flow MCP

The agent knows what the account can actually do

flow_inspect_account returns a live, normalized capability contract instead of relying on hardcoded assumptions:

• Image and video models, with stable IDs and the currently selected model
• Aspect ratios separated by media type
• Output counts separated by media type
• Selectable durations—only when Flow exposes a duration control
• Preview, original, available upscale, and unavailable/upgrade-only asset options
• Current Flow UI language and login state

Example from a real Spanish-language Flow account:

{
  "models": {
    "image": [
      { "id": "nano-banana-pro", "label": "Nano Banana Pro", "selected": false },
      { "id": "nano-banana-2", "label": "Nano Banana 2", "selected": true },
      { "id": "nano-banana-2-lite", "label": "Nano Banana 2 Lite", "selected": false }
    ],
    "video": [
      { "id": "omni-flash", "label": "Omni Flash", "selected": true },
      { "id": "veo-3-1-lite", "label": "Veo 3.1 - Lite", "selected": false },
      { "id": "veo-3-1-fast", "label": "Veo 3.1 - Fast", "selected": false },