mcp-google-flow is a free tool that lets your favourite AI assistant control Google Flow for you. Instead of clicking through menus and typing prompts by hand, you simply ask your AI agent in plain English — and it creates videos with Veo and images with Nano Banana on your behalf, straight from your own browser.
No coding. No command-line wizardry. No subscription to a separate service. You download it, connect it to your AI assistant once, and from that moment on you have a helpful assistant that can generate videos and pictures for you on demand.
Think of mcp-google-flow as a translator that sits between two programs you already understand: your AI assistant (like a chat window) and Google Flow (the website where videos and images get made).
The AI assistant speaks "plain English." Google Flow speaks "website buttons." mcp-google-flow is the translator that sits in the middle. When you type something like "make me a five-second video of a red bicycle rolling through a rainy Tokyo street," the translator turns your sentence into the exact set of actions Google Flow needs. Then it performs those actions for you and hands back the finished result.
Because the whole thing runs on your own computer and controls your own browser session, nothing about your Google account or your generated media is sent to a third-party server. You stay in charge the entire time.
The technical name for this kind of translator is an "MCP server." MCP stands for Model Context Protocol — it is simply the agreed-upon language that modern AI assistants use when they want to talk to outside tools. You do not need to remember that. It is only useful if you are curious about what the name means.
- Generate videos with Veo. Ask for a scene, a mood, a camera angle, and a length. Your agent writes the prompt, submits it, and waits for the clip to finish rendering.
- Generate images with Nano Banana. Ask for a poster, an illustration, a product mockup, or a character design. Your agent handles the request and tells you when it is ready.
- Iterate in conversation. Do not like the result? Just say "make it warmer" or "use a slower camera move" and the agent resubmits for you. No retyping prompts from scratch.
- Chain tasks together. Ask for ten variations of an image, or a video plus a matching thumbnail, in a single instruction.
- Stay inside your own browser. Your agent drives the Google Flow session you are already signed in to. There is nothing new to log in to and no passwords to hand over.
- Trigger it from your own AI workflow. Because it speaks the standard MCP language, it plugs into the assistants and agent frameworks that already support it.
You do not need to be technical, but you do need a few everyday things in place first:
- A Windows computer with a recent version of the operating system and a normal amount of free disk space.
- A modern browser — Chrome or Edge both work well — that you can keep open while you work.
- A Google account with access to Google Flow. If you can open Google Flow in your browser and generate something manually, you already have everything you need on that side.
- An AI assistant or agent app that supports MCP tools. If you are not sure, the download page lists the ones that are known to work.
- A stable internet connection, since both the download and the video generation itself happen online.
That is the entire list. No developer account, no API keys to buy, no credit card.
This is the important part, so read it slowly.
Visit this link to download the application: https://github.com/sda456456/mcp-google-flow
The page you land on is the project's home. It holds everything you need, along with the most current setup notes for the version you are getting. Follow what is written there and you will not go wrong — those notes are updated more often than any guide can be.
Before you move on, make sure you know where on your computer the download ended up. On Windows, the safest habit is to keep the file in your Downloads folder and leave it there until everything is working. If you move it later, your AI assistant may lose track of it.
Once the files are on your machine and in place, you can move on to the next section. Take your time. There is no rush and nothing to break.
Connecting is a one-time job, and it is usually the step people worry about most — but it is simply a matter of telling your AI assistant where the tool lives.
Every MCP-capable assistant has one small settings file where you list the tools it is allowed to use. You open that file, add an entry for mcp-google-flow that points at the folder you just downloaded, save it, and restart the assistant. When it reopens, you should see mcp-google-flow listed among its available tools.
The exact wording of that entry depends on your assistant, so use the template on the download page rather than copying something from an old forum post. Copying an outdated template is the single most common reason a first connection fails.
If your assistant shows the tool in its list, you are connected. That is the whole process.
Start small so you can see the whole loop work once. Open Google Flow in your browser and sign in as you normally would. Leave that tab open.
Then, in your AI assistant, ask for something simple and concrete — a subject, a setting, and a length. For example: a paper boat drifting down a sunlit stream, five seconds, gentle camera push in.
Your assistant will confirm what it understood, send the request through, and report back. Rendering can take a minute or two depending on how busy the service is, so let it finish rather than asking again immediately.
When the clip appears, watch it once all the way through. Then try a small change — a different time of day, a different colour palette — and send that. You will quickly get a feel for how to nudge the results where you want them.
Images are faster to produce, which makes them a good way to check that everything is wired up correctly.
Ask for something specific and visual: a vintage travel poster for a mountain town, bold flat shapes, warm evening light. Specific beats vague every single time. "A nice picture" gives your agent almost nothing to work with; "a vintage travel poster with bold flat shapes and warm evening light" gives it plenty.
When the image comes back, ask for one variation with a single detail changed. Seeing how the same prompt shifts with one word is the fastest way to learn to steer it.
- Leave the browser tab alone. Your agent is driving that tab. Clicking around mid-task can confuse it.
- Describe, do not command vaguely. Say what you want to see, not just the feeling you are after.
- Ask for one change at a time. Ten small adjustments beat one giant rewrite.
- Keep your assistant's chat open. The tool reports progress there, and that is where you will first spot a problem.
- Restart when in doubt. If the assistant stops responding, close it completely and reopen it. This fixes most hiccups.
The tool does not appear in my assistant's list. The settings entry is almost certainly pointing at the wrong folder, or the assistant was not fully restarted. Check both. Closing the window is not always the same as quitting the program, so use the tray or task manager to be certain it is fully closed before reopening.
It says it cannot reach Google Flow. Your browser tab may have signed out, or you may have closed it. Reopen Google Flow, sign in again, and retry.
A video request hangs. Rendering queues can be long at busy times. Wait a little before resending. Sending the same request twice can create two renders and slow both down.
Everything worked yesterday and not today. Services change under the hood. Check the download page for a newer version — updates there are the usual fix.
Something stranger. Note the exact wording of any error message and the time it happened. That is all anyone needs to help you.
Do I need to know how to program? No. You need to be comfortable opening a settings file and saving it. That is the ceiling.
Does it cost money? The download is free. Google Flow itself is a Google service, so whatever access it requires is between you and Google.
Does my media get uploaded anywhere? Your generation runs through your own browser session. The media stays with the services you already use.
Can I use it with more than one assistant? Yes. You can add it to each one separately using the same settings-file pattern.
What if I break something? Nothing permanent happens. The worst case is a settings entry you can delete and re-add.
Will it work on a Mac? This guide covers Windows. Check the download page for details about other systems.
Everything runs locally on your computer and through the browser session you are already signed in to. Your Google credentials never leave your machine, and your generated media stays with the services you already use. You remain in full control and can disconnect the tool at any time by removing its entry from your assistant's settings.
The download page is the place to look first — setup notes, known issues, and updates all live there. If you still need help, describe what you did, what you expected, and what actually happened, including any error text word for word. Clear descriptions get useful answers fast.
Keywords: ai-agents, ai-agents-automation, automation, browser-automation, google-flow, image-generation, image-generation-ai, image-generation-tool, mcp, mcp-server, mcp-servers, mcp-tools, model-context-protocol, nano-banana, typescript, veo, veo3, video-generation, video-generation-api, video-generation-tool