How to make videos with Claude or ChatGPT using MCP

Chat assistants like Claude and ChatGPT write scripts well, but they don’t render video by themselves. MCP (Model Context Protocol) closes that gap: it’s an open standard that lets an assistant call external tools. Connect a video generator’s MCP server, ask “make a 60-second explainer on how a CDN works”, and the assistant can write the script, show you the price, start the render once you agree, and hand you the finished file. This guide explains how it works, how to set it up, and how to prompt it well.

What MCP is, in plain words

MCP is a protocol that standardizes how AI applications connect to outside tools and data. A tool provider runs an MCP server that describes its tools (name, purpose, parameters). An MCP client (Claude, ChatGPT, Cursor, Claude Code, your own agent) reads those descriptions and decides when to call them during a conversation. The full specification is on modelcontextprotocol.io.

For video, that means the assistant does what it’s good at (understanding your request, writing, iterating with you) and the server does what the assistant can’t (voice synthesis, animation, rendering a 1080×1920 MP4).

What the assistant can do once connected

The Motiongist MCP server exposes the same steps as the web studio. A typical session follows this order:

1. Read the catalog (motiongist_catalog): voices, looks, lengths, the 10 narration languages, credit costs and your balance.
2. Create a project (motiongist_project_create) with your topic and settings. It also appears in your studio.
3. Write the script (motiongist_script_write), optional and free, or pass your own text.
4. Preview the narration (motiongist_narration_generate), optional and free: an audio link to hear the voice on your script.
5. Review the storyboard (motiongist_storyboard_write, then motiongist_storyboard_get), optional and free: the scene-by-scene plan, which you can edit with motiongist_storyboard_update.
6. Estimate the price (motiongist_estimate): the exact credit price and a signed quote valid for one hour.
7. Generate (motiongist_generate), only with your approval and the unchanged quote.
8. Track and fetch (motiongist_job_get, then motiongist_video_get): progress, then signed download links for the MP4, the SRT captions, a contact sheet and the storyboard.

The approval rule

An assistant should never spend your money silently. That’s why generation requires two things: approved set to true, meaning you accepted the price, and the signed quote returned by the estimate, unchanged. If the project changed after the estimate (a new script, a different voice or length), the quote is refused and the assistant has to estimate again and ask you again. The render can’t cost more than the quote, and failed or cancelled generations are refunded.

How to connect it

The server lives at https://motiongist.com/api/mcp and uses the Streamable HTTP transport. MCP access is included from the Studio plan; on other plans, only the catalog tool answers. There are two ways to authenticate:

• OAuth (recommended for Claude.ai and ChatGPT): you add the URL, the client opens a Motiongist sign-in page, you pick the workspace and approve. No key to copy.
• An MCP key (Claude Code, Cursor, custom agents): create a key in your account’s developer settings and send it in the x-api-key header.

Claude (claude.ai and Claude Desktop)

1. Open Settings → Connectors → Add custom connector.
2. Name it Motiongist and paste the server URL.
3. Click Connect, sign in and allow access, then enable the connector in your conversation.

ChatGPT

ChatGPT’s support for custom MCP connectors depends on your plan and on settings OpenAI changes over time; as of October 2026 it goes through a developer mode in the app’s connector settings. The general steps: enable developer mode, create a connector with the server URL and OAuth authentication, then sign in. Check OpenAI’s current documentation if the menus look different.

Claude Code, Cursor and other clients

In Claude Code, add the server from the terminal with the HTTP transport and your key as a header (claude mcp add --transport http motiongist followed by the URL and --header "x-api-key: …"), or skip the header and run /mcp to sign in with OAuth. Cursor, VS Code, Windsurf and custom agents accept a standard MCP server configuration with the URL and the header. The exact snippets are in the MCP docs.

Prompts that work

The assistant fills the project settings from your request. The more your prompt reads like a brief, the less back-and-forth you’ll need. Some examples you can adapt:

“Make a 60-second vertical explainer on how a CDN works, for beginners. Use the Night look and George’s voice. Show me the script and the price before generating.”

“Here’s our release note [paste]. Turn it into a 90-second video for existing customers, in Spanish, with Sarah’s voice. Let me review the storyboard first.”

“Give me five video ideas for a personal finance channel, then write a 45-second script for the best one and estimate it in 3D.”

Good habits when prompting:

• State the length: short (45–60 s), standard (90–120 s) or long (150–180 s). It drives both the script and the price.
• Name the audience: beginners, decision-makers, developers. It changes the vocabulary.
• Ask for the storyboard when the topic is precise or technical. It’s free and it’s the cheapest place to catch a mistake.
• Ask the assistant to fact-check the script against your source before estimating. You’re the one publishing.
• Say the language explicitly if it’s not English: Spanish, Portuguese, French, German, Italian, Dutch, Polish, Turkish or Indonesian.

Where MCP shines, and where it doesn’t

MCP is ideal when the video is part of a bigger conversation: you’re already working on a launch, a lesson or an article with your assistant, and the video is one more output. It also works well for small batches, like “make one short per section of this document”, where the assistant handles the repetition.

For fully automated pipelines that run without a human in the chat (a nightly job, a CMS hook), a direct integration with the REST API is more predictable: you control retries, webhooks and logging in code. See video generation API for developers. And for a single video where you want to see every setting, the web studio is still the fastest.

Troubleshooting

A quick checklist before your first video

1. Your workspace is on a plan with MCP access, with enough credits for the length you want (see the calculator).
2. The connector is added and enabled in the conversation.
3. Your prompt names the topic, audience, length, language and, ideally, a voice from the voice library.
4. You read the script (and the storyboard, for technical topics) before approving the price.
5. You check the final video before publishing it, as you would with any content.

Once it’s set up, making a narrated explainer becomes one more thing your assistant can do mid-conversation. To make the most of it, the fundamentals don’t change: a sharp hook, one idea per scene, and a script written for the ear (script template).