Roll Camera From Chat: How the OpenClips MCP Door Works
MCP lets an AI assistant call the studio as a tool. Here is how the OpenClips MCP door is built — what the agent does, how the connection authorises, and why the app can never start it for you.
Scene select — 6 scenes
The studio has a second door. The composer is the one you press with your hands; MCP is the one an agent walks through. Same models, same credits, same library at the end of it — a different way to call action. Here is how that door is built.
What MCP actually does
MCP — the Model Context Protocol — is an open standard that lets an AI assistant call external tools as if they were built in. Point a client at an MCP server once and the assistant gains that server’s capabilities natively: no copy-paste between tabs, no browser automation, no glue script you maintain.
For a video studio, that changes the shape of the work rather than the speed of it. You stop describing a render and start describing an outcome.
What the agent does with it
Three moves, in order:
- Say it in plain words. “A 15-second vertical testimonial for my coffee brand, natural light.” That is the whole brief.
- The agent drafts the shot list. Model, duration, ratio, references, and the quote for the render it is about to place.
- The cut lands in your library. Download it, remix it, or ship it — the same place a composer render arrives.
Three briefs worth trying
These are conversation instructions, not render prompts — the agent turns each one into the shot list.
Cut three 15s UGC variants for my coffee brand — one bold-claim hook, one demo-first, one callout. Vertical. Tell me the quote for all three before you place them.
Render a vertical product orbit for TikTok, then render the same beat again as a 16 cut for YouTube. Do not crop the first one.
List the video models I can render on, check my balance, then queue the shot on whichever flagship I can afford to run twice.
The chain an agent is actually good at
A single render is the least interesting thing to hand an agent. A chain is the interesting one: lock a character or a product as a still, block the beats as keyframes, then animate the frame that won. Each step consumes the step before it, which is exactly the kind of bookkeeping a conversation is good at and a human is bored by.
How the connection authorises
The design here is deliberately one-directional, and it is worth knowing before you wire anything up.
- The client starts it. Your MCP client sends you to the OpenClips connect page in a browser.
- You approve it there. That page mints a short-lived code and hands it back to the client’s own callback.
- The code is spent once, by the backend, for a 30-day personal access token.
- OpenClips can only finish a flow that began elsewhere. There is no Connect button in the app, by design.
Where this is going
The composer is for hands. MCP is for agents. Same models, same balance, same library — and the interesting work is the part neither door does alone: you decide what the ad has to say, and the machine does the shot list.
