---
name: cueframe
version: 1.0.0
description: Use when a task needs to turn footage into finished video — compose or edit a video, cut a 9:16 short or reel from a longer clip, add captions and motion graphics, reframe to another aspect ratio, or render production-quality MP4. CueFrame is the video-composition engine your agent drives; you bring the intent, it does the editing and rendering.
---

# CueFrame — set up & drive video from your agent

CueFrame is an API-first, agent-native video **compose** substrate. You bring the
creative intent; CueFrame ingests footage, composes the edit as data, and renders
production-quality video. Content-aware by default: it detects faces, transcribes
speech, reframes around the active speaker, and wires captions from the transcript.


## Pick the job, then its skill

Fetch the skill for the job at `https://api.cueframe.ai/v1/skills/<name>`:

| You were asked to… | Skill |
|---|---|
| Make a reel/short/vertical clip from footage | `make-a-social-reel` |
| Make a brand-identity video | `brand-reel` |
| Clip a podcast/talk into shorts | `clip-a-talking-head` |
| Make a launch/feature announcement video | `launch-video` |
| Ship the same edit to every platform format | `every-format-from-one-edit` |
| Add or fix music/sound | `add-music-bed` |
| Restyle a video to a (new) brand | `rebrand-a-video` |

Whatever the job: `video-craft-standards` defines what DONE means — read it
before composing, score against it before rendering.

## Setup (one line, no API keys)

```bash
npx -y cueframe install
```

Idempotent. It detects every agent on this machine — Claude Code, Cursor, Codex,
VS Code, Windsurf, Zed, Gemini CLI, Copilot CLI and more — and wires the CueFrame
MCP server into each, then installs the CueFrame agent skills. **No API key is
pasted.** On your first call a browser opens and you click **Allow** (sign-up
happens there if you're new). Credentials stay local; revoke anytime in the
console.

Claude Desktop and ChatGPT are not wired by the installer: neither reads a local
MCP config the CLI can write (Claude Desktop speaks stdio only; ChatGPT has no
local config at all). Both use the hosted connector below instead.

Restart your agent session after install to load the MCP server and skills.

## Remote MCP server (if you can't run a shell)

Add this hosted server as a custom connector — Streamable HTTP, OAuth 2.1:

```
https://api.cueframe.ai/v1/mcp
```

## Using it

Once wired, the tools are exposed as `mcp__cueframe__*`. Orient first with
`get_account` — read `balance`, not `included`: rendering, generation, previews
and the judge all spend one shared credit wallet, so a positive balance means go.
Then the core loop:

1. **Ingest** — `create_project`, then `import_media` (a URL) or `generate_media`
   (a prompt); `get_media_context` returns detected faces + transcript.
2. **Compose** — author the edit and run the Director (`compose`), or cut a short
   from a clip suggestion (`suggest_briefs` → `compose`).
3. **Render** — `create_render` → an MP4 `downloadUrl`; re-render to any aspect
   ratio from the same composition.

Prefer the terminal? The same thing runs over the `cueframe` CLI
(`cueframe login` for browser auth) or raw REST at `https://api.cueframe.ai/v1`.

## Free to start

You start with a one-time free credit grant — no card — and it funds real work:
rendering, AI generation, preview stills and the judge all draw on it. Check
GET /v1/me (or the MCP `get_account` tool) for your live balance and per-call
ceilings; brand kits, stock imports and dry-run validation cost nothing. Agents
can also pay per call with x402, no account required.