Hand it toan agent.
Cutlist speaks MCP, so any agent runtime can cut video as a tool call. The same engine the studio uses, the same credits, the same output - just driven by something that is not a person clicking.
{
"mcpServers": {
"cutlist": {
"url": "https://cutlist.app/api/mcp",
"headers": {
"Authorization": "Bearer clk_live_..."
}
}
}
}JSON-RPC 2.0 over a single POST. Open the URL in a browser and it describes itself.
cut_video
Point it at a video URL and it finds the moments, reframes them, captions them and writes the post text. Optionally hand it timestamps you already have.
get_clips
Poll a job. Returns each clip with its score, what the model saw and heard, a download URL and the social caption.
estimate_cost
What a job would cost in credits and what the balance is, before anything is spent.
list_jobs
Recent jobs on the project, newest first.
Things worthwiring up.
A weekly show that clips itself
An agent watches for a new episode, calls cut_video, waits, then drafts the week's posts from the captions it gets back.
Clip the moment it happens
A stats platform logs a scoring play and hands the timestamp straight to cut_video. Detection is skipped, so the clip bills at the anchored rate and lands in about the time it takes to render.
A researcher's highlight reel
An agent working through hours of interview footage pulls the passages that matter and returns them as clips with alt text, rather than a wall of transcript.
Budgeted batch work
estimate_cost first, then cut only what fits. An agent can stay inside a credit budget without a human watching it.
An agent spends real credits, so cut_video reports the hold it took and estimate_cost exists to ask the price first. Nothing runs up a bill quietly.
MCP is on every plan, the trial included - it is the same engine and the same credits as the studio, so there is nothing extra to buy. Make an account, then ask us for the project key: the studio has no key screen yet, so that one step is still a person on our end rather than a button.