Capabilities — Cue
Surfaces
| Surface | Identity |
|---|---|
| MCP | io.github.SylphxAI/cue over stdio, npx -y @sylphx/cue |
| CLI | cue |
| SDK | @sylphx/cue/sdk |
Owned capabilities
| Capability | Tool | Evidence |
|---|---|---|
| Timeline document | read_video | default: streams, chapters, embedded subtitles. quality or include_scenes adds ffmpeg scene boundaries. The shipped server does not extract keyframes |
| Timestamped search | search_video | embedded subtitle matches with timestamps. No speech recognition |
| Frame evidence | video_evidence (render_frame, crop_frame, ocr_frame) | a separate follow-up: frame, crop, and OCR at a known timestamp |
Evidence contract
Every result carries timestamp_ms, stream or subtitle locators, source hash, route, warnings and gaps. A frame locator is present only after a frame follow-up. See EVIDENCE_CONTRACT.md.
Not owned
Cloud video or speech-recognition APIs as a requirement, per-frame vision summaries, media-library management, and still-image semantics.