REST, MCP and CLI

Three calls to finished video.

Upload footage, write the edit as data, render. Your agent drives it over MCP; your pipeline calls the same API directly.

Get an API keyAPI reference

Upload

POST your footage, a file or a URL. CueFrame transcribes the audio, finds the speakers, and hands back media your edit can reference.

Compose

Write the edit as data: which clips, where the captions sit, your brand. Validating an edit is free, so agents can check before they spend.

Render

One call turns the edit into finished video, any aspect ratio up to 4K. The same edit renders the same video, every time.

One command wires your agent.

npx -y cueframe install

Detects Claude Code, Cursor, Codex, VS Code and 10 more on your machine and wires CueFrame into each. No API key: your agent opens a browser to authorize on first call. On Claude Code you can instead paste its skill; on ChatGPT, add api.cueframe.ai/v1/mcp as a connector.

MCP

Every CueFrame tool, discoverable and callable by your agent. Hosted at api.cueframe.ai/v1/mcp, so ChatGPT and Claude connect to it as a remote server.

REST

A plain HTTP API with API keys for pipelines and backends. Everything the agents can do, your code can do.

CLI

cueframe on your machine: import footage, author edits, render, and eject a composition into code when you want to own it.

The edit is data.

This is a real op from the composition format: the same shape the desktop editor writes and the renderer reads. Change one line and only that one thing changes; pin the edit and the same video comes back every time. Validating a composition is free, so agents check before they spend.

{
  "type": "clip.add",
  "clip": {
    "id": "l1",
    "startTime": 15,
    "duration": 5,
    "source": {
      "kind": "overlay",
      "primitiveId": "lower-third",
      "params": {
        "text": "Daniel Marsh",
        "subtitle": "Host",
        "tokens": { "accentColor": "#f7c948" }
      }
    }
  }
}

Change one line, not the whole video.

The edit is plain data. Read it, version it, change one line. A change costs a re-render, not a re-edit. Run it again untouched and the same video comes back. That's the point.

format: "9:16"
format: "1:1"

A new shape for a new platform. Same footage, same type, same timing. Nothing was re-cut.

overlay: "48 hours"
overlay: "72 hours"

The recipe changed, so the number did. The shot behind it never moved.

Checking an edit is free. Previews cost pennies. You pay when you render, and a short costs $0.50. Next week's episode reuses the same edit, so you never start over. And it exports to Premiere or Final Cut whenever you want.

Your agent gets the same AI.

What it saw, as data

Every upload comes back with a word-timed transcript, speaker labels, and what's in the frame. Your agent composes with what CueFrame saw, not just file names.

It can score its own cut

Score a cut and get the weakest thing back. The compose loop is agent-driven: render, look, score, fix, repeat until it's good.

Footage on request

No source material? Agents can search licensed stock from inside the API and cut it like any upload. Searching is free.

Brand from a URL

Point CueFrame at a site and it extracts the brand kit: colors, fonts, logo. Every render after that stays on brand.

Start with the docs.

The setup guide gets an agent composing in minutes; the API reference covers every call. Free credits are on the account, no card.

Get an API keyAPI referenceGitHub