Skip to main content
Glama

🎬 Google Flow MCP Server (google-flow-mcp)

Use Google Flow (Google's AI video studio at flow.google.com) from Claude Desktop, Claude Code, ChatGPT, and Gemini CLI.

Plan a shot with your assistant, then ask it to generate the video. The server drives Flow with your own Google account and Flow credits, so you don't need an API key.

⚠️ Unofficial project. This is not made by or affiliated with Google. Sign-in mode automates the Flow website in a Chrome window on your computer. Automating a website may go against Google's Terms of Service, and Google can change Flow's UI at any time, which can break this tool. Use it at your own risk with your own account.


✨ Tools

Tool

Needs

What it does

flow_browser_login

Chrome

Opens Flow in Chrome so you can sign in once. Your sign-in is saved in ~/.google-flow-mcp/chrome-profile.

flow_text_to_video

Flow sign-in

Starts a video from a text prompt with your Flow credits and returns a job_id right away. Video tools run in the background because a clip takes 1-5 minutes, longer than ChatGPT waits for a tool.

flow_check_video

nothing

Gets a video started by any video tool (pass its job_id) and saves the MP4. Waits up to 45s per call; call again while it is still generating. Free, no credits.

flow_frames_to_video

Flow sign-in (experimental)

Animates a starting image, or transitions from a first frame to a last frame.

flow_reference_to_video

Flow sign-in (experimental)

Generates a video guided by reference images (product, character, style).

flow_generate_frame

Flow sign-in

Generates an image (Nano Banana) with your Flow credits, saves it at full size, and returns the picture so Claude and ChatGPT can show it. Sets the aspect ratio (16:9, 4:3, 1:1, 3:4, 9:16) in Flow's settings and makes one image per request. Reference images are experimental.

flow_show_latest

nothing

Shows the most recent saved image or video again. Free, no credits.

flow_extend_or_edit_video

Gemini API key

Extends or edits an existing clip.

flow_storyboard_director

Gemini API key

Turns a brief into a multi-shot storyboard with frame and motion prompts.

The video and image tools use your Flow sign-in by default. If you set GEMINI_API_KEY and pass use_browser_credits=false, they use the Gemini API instead. In sign-in mode Flow chooses the image size (for example 768×1376 for 9:16), so image_size and model only apply with an API key.

How sign-in mode works

  1. Opens Flow in Chrome with a saved profile, so you stay signed in.

  2. Creates a new project, types your prompt, and starts the generation.

  3. Flow asks before spending credits (for example "costing 30 credits?"). The tool clicks Approve for that one generation only, never "Always approve", and returns the cost as credits_cost. The tools also read your credit balance before and after, and report credits_used, credits_before and credits_after.

  4. Waits for the result (up to 10 minutes for video, 4 for images), then saves it to your media folder. Images are saved at full size, not Flow's small preview. If the page misses the finished image, the tool reloads the project to find it.

Runs in the background. On a Mac the tool's Chrome is hidden (like Cmd+H) and minimized as soon as it starts, so it stays signed in without showing a window. Only its Dock icon appears while it works. Set FLOW_SHOW_BROWSER=1 to watch it work. Only flow_browser_login shows a window, so you can sign in.


Related MCP server: gflow-cli

🚀 Install (2 minutes)

You need Google Chrome, a Google account with Google Flow credits, and uv (a Python installer). Install uv once:

curl -LsSf https://astral.sh/uv/install.sh | sh

No cloning or Python setup needed: every command below runs the server straight from GitHub with uvx.

The first time you use it, ask your assistant to run flow_browser_login and sign in to Google in the Chrome window that opens. You only do this once.

Generated videos and images are saved to ~/Downloads/google-flow.


🔌 Claude Code

claude mcp add google-flow -- uvx --from git+https://github.com/abdureda1999-ai/google-flow-mcp google-flow-mcp

🔌 Claude Desktop

Find the full path of uvx with which uvx (usually ~/.local/bin/uvx), then add this to ~/Library/Application Support/Claude/claude_desktop_config.json and restart Claude:

{
  "mcpServers": {
    "google-flow": {
      "command": "/Users/YOU/.local/bin/uvx",
      "args": ["--from", "git+https://github.com/abdureda1999-ai/google-flow-mcp", "google-flow-mcp"]
    }
  }
}

♊ Gemini CLI

Add this to ~/.gemini/settings.json, restart gemini, and run /mcp to check that google-flow is listed:

{
  "mcpServers": {
    "google-flow": {
      "command": "uvx",
      "args": ["--from", "git+https://github.com/abdureda1999-ai/google-flow-mcp", "google-flow-mcp"]
    }
  }
}

The Gemini web and mobile apps can't connect to custom MCP servers like this one, so use Gemini CLI.

🤖 ChatGPT

ChatGPT runs in the cloud, so it needs a public link to the server on your computer. One command sets that up. First install Cloudflare's free tunnel tool once:

brew install cloudflared

Then start ChatGPT mode:

uvx --from git+https://github.com/abdureda1999-ai/google-flow-mcp google-flow-mcp chatgpt

It prints a link like https://something.trycloudflare.com/<secret>/mcp. Then add it to ChatGPT once:

  1. In ChatGPT, open Plugins (Settings → Plugins → Browse directory), click Add → Add custom MCP server.

  2. Fill in:

    • Name: Google Flow

    • Server URL: the link from your Terminal

    • Authentication: No authentication

  3. Tick I understand and want to continue, then click Create as a plugin.

Using it: in any chat, type @Google Flow, pick it, and ask for an image, for example:

@Google Flow make a 9:16 worm's-eye shot of a red soda can on grass against a blue summer sky

ChatGPT shows "Generating image in Google Flow…", and after about a minute the picture appears right in the chat with a download link. Videos come back with a download link.

  • Use a normal chat. ChatGPT's Work / agent mode opens Flow in its own browser instead of using this plugin, and won't show the image in the chat.

  • Keep that Terminal window open while you use ChatGPT. Your computer must stay on.

  • The free Cloudflare link changes each time the tunnel restarts, and ChatGPT can't edit a plugin's link, so you'd delete and re-add the plugin. To avoid that, use a permanent link (below).

  • Click Refresh tools in Plugins → Google Flow after updating this server.

A free ngrok account gives you one domain that never changes:

  1. Sign up at dashboard.ngrok.com, then run the ngrok config add-authtoken ... command shown under Your Authtoken.

  2. Copy your free domain from Domains (like something.ngrok-free.app).

  3. Start ChatGPT mode with it:

brew install ngrok
FLOW_NGROK_DOMAIN=something.ngrok-free.app uvx --from git+https://github.com/abdureda1999-ai/google-flow-mcp google-flow-mcp chatgpt

The secret part of the link is saved in ~/.google-flow-mcp/secret, so the full link stays the same every time. Add it to ChatGPT once. If ngrok drops (sleep, Wi-Fi change), the server reconnects on the same domain.

Running it as a background service (launchd)? Add FLOW_NGROK_DOMAIN to the service's EnvironmentVariables and stop any separate cloudflared service, so only one tunnel points at the server.

Switching an existing ChatGPT plugin to the permanent link: ChatGPT can't edit a plugin's URL. In Plugins → Google Flow, click Uninstall, then Delete app (uninstalling alone keeps the old app, and its name blocks the new one with "An app with this name already exists"). Then Add → Add custom MCP server with the new link and No authentication.

  • 🔒 Anyone with the link can spend your Flow credits. Don't share it, and press Ctrl-C when you're done.

uvx --from git+https://github.com/abdureda1999-ai/google-flow-mcp google-flow-mcp --transport openapi --port 8000
cloudflared tunnel --url http://127.0.0.1:8000

In Create a GPT → Configure → Actions → Import from URL, paste https://YOUR-TUNNEL.trycloudflare.com/openapi.json. This endpoint has no secret, so only run it while you're using it.


🧑‍💻 Developing locally

git clone https://github.com/abdureda1999-ai/google-flow-mcp.git
cd google-flow-mcp
python3 -m venv .venv && .venv/bin/pip install -e .
.venv/bin/google-flow-mcp --help

💡 Example prompts

  • "Use flow_text_to_video to make a cinematic studio commercial of a matte black sports car under moving neon light, 16:9."

  • "Make a 9:16 vertical clip of coffee being poured in slow motion with warm morning light."

  • "Use flow_generate_frame to make a 9:16 worm's-eye product shot of a cold soda can against a blue summer sky."

  • "Use flow_frames_to_video with /path/to/start.png as the first frame and /path/to/end.png as the last frame for an 8-second transition."


⚙️ Configuration

Variable

Default

Purpose

FLOW_MEDIA_DIR

~/Downloads/google-flow

Where videos, images and screenshots are saved.

GEMINI_API_KEY

not set

Turns on the API-only tools.

FLOW_HOST / FLOW_PORT

127.0.0.1 / 8000

Address for the HTTP transports.

FLOW_SECRET

saved in ~/.google-flow-mcp/secret

Secret part of the ChatGPT link.

FLOW_NGROK_DOMAIN

not set

Your free permanent ngrok domain, so the ChatGPT link never changes.

FLOW_SHOW_BROWSER

not set

Set to 1 to show the Chrome window while generating.

FLOW_PUBLIC_URL

not set

Your own public address (for example a named Cloudflare tunnel) so the ChatGPT link stays the same. With it, no quick tunnel is started.

🛠 Troubleshooting

  • Use a different Google account for Flow: move ~/.google-flow-mcp/chrome-profile aside (for example rename it to chrome-profile-backup), then run flow_browser_login (or .venv/bin/python examples/sign_in.py) and sign in with the other account. Results include flow_account, and Flow links carry authuser= so they open in that account.

  • ChatGPT hangs when the server runs as a background service (launchd) on a Mac: macOS blocks background apps from reading ~/Downloads until you approve it. Set FLOW_MEDIA_DIR to a folder outside Downloads, Documents and Desktop, for example ~/Google Flow.

  • ChatGPT says "Google Flow's connection failed": ChatGPT can't reach the link saved in the plugin, and restarting the server won't fix that. A free Cloudflare quick tunnel can expire (tunnel.log shows Unauthorized: Tunnel not found), which gives you a new link. Check the URL in Plugins → Google Flow matches the one the server printed, and switch to a permanent link.

  • A video "timed out" in ChatGPT: update this server. Video tools now return a job_id at once, and ChatGPT fetches the clip with flow_check_video. Click Refresh tools in the plugin's settings after updating.

  • ngrok ERR_NGROK_105: the saved authtoken isn't valid. Run the full ngrok config add-authtoken ... command from your ngrok dashboard, with the real token, not a placeholder.

  • Flow in another language: the tool always opens Flow in English (hl=en), so accounts set to Arabic or other languages work.

  • "Still on the Google Flow landing page": run flow_browser_login and finish signing in.

  • "Could not find the Flow prompt box": Flow's website changed. Please open an issue.

  • Chrome says the profile is in use: close any Chrome window the tool left open, then try again.

  • not_started status: Flow didn't ask to approve a generation. Check the saved ui_screenshot.

📄 License

MIT

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      A
      quality
      A
      maintenance
      Drives Google Flow from an agent or the terminal: Veo text-to-video, image-to-video and clip extension, plus Imagen image generation. Runs locally against your own Google account with Flow access; video generation bills that account's credits.
      20
      400
      MIT
    • A
      license
      B
      quality
      B
      maintenance
      Enables users to generate images and videos on Google Flow through browser automation of an already signed-in Chrome session, downloading the finished media to disk without any API keys. It drives Flow's agent with a range of image and video models, plus reference images, first/last frames, and ingredient-based video jobs, while reusing the user's own Google AI Pro credits.
      17
      MIT