How to use Claude or ChatGPT to create a product video
TL;DR
- On their own, Claude and ChatGPT write the script and storyboard, and you build the video somewhere else.
- Through an MCP server, the same assistant creates the project, adds narration, previews frames, and exports.
- Give the agent a template, your source material, and a recording, then ask it to review its own frames before export.
- Use the agent for batches and multi-tool workflows, and do one-off trims on the timeline yourself.
Ask an AI assistant how to make a product video with Claude or ChatGPT and you get the same answer almost every time. The assistant writes the script and the storyboard, and then you carry them to a separate video tool to render. That answer is out of date. Connect Claude or ChatGPT to a video tool through an MCP server and the assistant can create the project, generate the narration, place the callouts, and export the finished video while you review it.
The customer education team at Teamworks already works this way. Most of its videos start in Claude: someone names the template, uploads the raw screen recording, and lists the edits and callouts they want. The team opens the video editor only at the end for final touches, and a video that took 4 to 5 hours now takes 20 to 30 minutes.
Can Claude or ChatGPT make a product video on their own?
Not a narrated walkthrough of your software, no. Neither assistant can record your product or assemble a finished video of it by itself. Ask ChatGPT and it says so: the assistant handles the concept and the script, and "a separate video tool" generates "the actual moving footage."
An MCP server is what lets the assistant do the building. Model Context Protocol (MCP) is an open standard that lets an AI assistant call actions inside another tool. A video MCP server exposes actions such as creating a project, adding a clip, generating a voiceover, and exporting. The assistant decides which actions to call and in what order, and the video tool does the rendering.
That gives you two routes to the same video.
What differs | Route 1: plan in chat, build elsewhere | Route 2: the agent drives the video tool |
|---|---|---|
What the assistant produces | A script, a storyboard, voiceover copy | A project you can open, edit, and export |
Who assembles the video | You, in an editor | The agent, through MCP |
What an edit costs | Reopen the editor and redo it by hand | One instruction in chat |
How do you make a product video with Claude or ChatGPT without connecting anything?
Route 1 works, and it's where most teams start. The assistant only writes here, and you record, edit, and export in your own tool.
- Give it the source, not the topic. Paste the release notes or PRD. "Make a video about our new export feature" invites the assistant to guess at buttons that don't exist.
- Ask for a scene list. Each scene needs a narration line and the on-screen action, which becomes your recording plan.
- Record the product. Follow the scene list in one take and narrate roughly.
- Rewrite the script from the recording. Paste the transcript back and ask for narration that matches what the product actually did. Skip this and your voiceover describes a menu the viewer never sees.
- Assemble it in your editor. Import the recording, add the voiceover, zooms, and captions.
The weakness shows up the week after. The assistant that wrote the script can't see the video, so every change happens by hand, and the next release sends you back to step 3.
How do you connect Claude to a video tool through MCP?
We'll use the Clueso MCP as the worked example, because it's the one we build and it works with both assistants. The steps carry over to any video MCP server that exposes editing actions, and our guide to what a video MCP server actually does shows how to check for that.
Claude connectors need a paid Claude plan (Pro, Team, or Enterprise).
- Open Claude on the web or desktop, click your profile, and go to Settings.
- Open Connectors and click Add custom connector.
- Enter
https://connect.clueso.io/mcp. - Sign in to Clueso in the browser window that opens and click Allow.
Clueso now appears in the tools panel for every new chat. In Claude Code, run claude mcp add --transport http Clueso <https://connect.clueso.io/mcp> instead. You don't need a Clueso account first, because one is created when you sign in.
How do you connect ChatGPT to a video tool through MCP?
ChatGPT connects through a plugin. On ChatGPT Business, Enterprise, or Edu, a workspace admin may need to enable it first.
- Open the Clueso plugin in ChatGPT, or go to Plugins and search for "Clueso".
- Click Connect, sign in to Clueso, and click Allow.
- In any chat, type @Clueso, or click the + below the chat box and pick Clueso.
If your admin hasn't enabled the plugin, connect in Developer mode instead: Settings → Apps, turn on Developer Mode, click Create, enter https://connect.clueso.io/mcp, name it "Clueso", and authorize.
To test either connection, ask "Show me my most recently updated Clueso project" and approve the tool call.
What prompts should you give the agent to build the video?
Route 2 runs in five steps. Each prompt below builds a feature release video, so swap in your own feature and files.
1. Confirm the workspace. A connection often reaches more than one workspace.
Which Clueso workspaces can you see? Use the Customer Education workspace for everything in this chat.
2. Start from a template. A template fixes layout and branding before the agent makes a single design decision. In Clueso these are Clueprints, and the agent can search your workspace and the community library for one.
Find a clueprint for a feature release video. Show me the best two matches and why they fit before you build anything.
3. Hand over the source and the recording in one message. Say who the video is for and how long it should run.
Build a 60-second feature release video for our new bulk export in Reports, using the clueprint we picked. The release notes and a raw screen recording are attached. The audience is admins who already know Reports. Write the narration from what the recording shows, generate the voiceover, sync it to the on-screen steps, and add a zoom on the Export button.
Behind that message, the agent creates the project, adds the recording as a clip, generates AI voiceover, syncs narration to the screen, and layers in zooms and callouts.
4. Make it check its own work. The agent can render a preview frame of any clip and look at it.
Render a preview of every clip. Flag any text that overlaps the UI, any zoom that lands before the narration mentions the button, and any clip that runs longer than its voiceover.
5. Export with what you need. Clueso's export defaults to 1080p at 30fps with no captions, so ask for anything else explicitly.
Export the video with burned-in captions, and write a short step-by-step guide from the same project for the help center.
Because the step-by-step article comes from the same project, it describes the same screens as the video.
When should you use an AI agent instead of the video tool's own interface?
Connect the agent when the work repeats or crosses tools, and keep small visual changes in the editor.
Use the agent through MCP when | Use the editor's own interface when |
|---|---|
The video starts in another tool, like release notes in Linear that should become a release video sent to Slack | You're making a one-off trim or nudging an element you can see |
You need versions: one video per persona, per language, or per aspect ratio | You're judging the feel of a cut, which is faster to watch than to describe |
A change touches many videos, like new brand colors across a whole library | A reviewer needs to comment on a specific frame |
Bulk edits save the most time: one rebrand prompt can update every project in a workspace and re-export the affected videos.
Recording usually still needs a person, because the agent edits footage you give it. On the Enterprise plan, Clueso Agents can record your product too, and the assistant can start that recording through the MCP.
What security teams usually ask before approving a video MCP?
Enterprise buyers tell us their security teams are "pretty strict on the approvals of MCPs." Have these answers ready before you ask.
- How does the agent sign in? Through OAuth 2.1 with PKCE. There are no API keys, and every token belongs to a named person and the workspaces they can already reach.
- What can it touch? Only the projects and assets the signed-in account can access. Video content stays on Clueso's infrastructure and isn't used to train external AI models.
- How do we turn it off? Revoke access in the Clueso workspace settings, and the agent's next call fails.
- Can we inspect it first? Yes. The server returns its tool list without a token, so a reviewer can read every action the agent could take before anyone grants access.
Frequently asked questions
Can Claude make videos?
Not by itself. Claude writes the script and storyboard, but it needs a video tool to build and render the file. Connected to a video MCP server, Claude can create the project, generate the voiceover, and export the video from the chat.
Can ChatGPT make videos?
ChatGPT plans videos well and recommends a separate tool for the footage. For a video of your software, connect it to a video tool through MCP so it builds the project instead of handing you a script.
How do I connect an AI agent to a video creation tool so it can make how-to videos automatically?
Add the video tool's MCP server to your assistant, sign in once, and give the agent a template and your source material. In Claude that's a custom connector, and in ChatGPT it's a plugin.
What should you try first?
Pick the most recent feature you released without a video. Connect the MCP, paste the step 3 prompt with that feature's release notes and a 2-minute recording, and time it from first message to export. Teamworks went from 4 to 5 hours a video to 20 to 30 minutes. Your own number tells you how many of next quarter's releases can ship with one.
Ashish Upadhyay
Senior Content Marketing Manager
Ashish is a Senior Content Marketing Manager at Clueso with 7+ years of experience across content, product, brand marketing. Now his mission is to help product and customer education teams realize the value of video-based learning. Outside of work, Ashish sketches, sings, plays the guitar, cooks, and does all things LLMs can't yet.
