Cue

Name: Cue
Author: SylphxAI

Local

sylphxai.github.ioMedia & Design3Updated 5w ago

Repository Visit website

About Cue

Cue is a local video MCP server for agents that need searchable, timestamped evidence. Its default read returns container, stream, chapter, and embedded subtitle data; scenes and frames are requested separately.

Key features

01Read video metadata, streams, chapters, and subtitles
02Search embedded subtitles with timestamps
03Render, crop, or OCR a frame by timestamp
04Optionally detect scene boundaries
05Include source hashes and subtitle locators

Use cases

01Find a specific quote in a video
02Summarize meeting videos with chapters or subtitles
03Locate a frame showing code in a demo
04Find scene changes in a demo video

MCP support

Cue exposes the read_video, search_video, and video_evidence tools to MCP clients for reading local videos, searching subtitles, and retrieving frame evidence at a specified timestamp.

Configuration

Local server config is the same for every client. Copy it into your client’s MCP config file.

mcp.json · Universal config

{
  "mcpServers": {
    "sylphxai-cue": {
      "command": "npx",
      "args": [
        "-y",
        "@sylphx/cue"
      ]
    }
  }
}

Run commandnpx -y @sylphx/cue

Local servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.

Packages

npm@sylphx/cue0.3.5stdio

Tools

No tool list yet

Local servers expose tools only once running on your machine, and the maintainer has not provided a tool list.

Details

From the project README

Merged into anymd (2026-09-25). anymd reads PDFs, Office files, EPUB, web pages, images and video into clean Markdown for AI agents: npx -y @sylphx/anymd. This repository is archived.

Cue

Video answers with timestamp-level proof

Cue gives agents a local video timeline they can search and cite. The default read returns container metadata, streams, chapters, and embedded subtitles. Scenes and frames are separate.

npx -y @sylphx/cue

For Claude Code:

claude mcp add cue -- npx -y @sylphx/cue

Read a local video

{
  "sources": [{ "path": "/absolute/path/to/demo.mp4" }]
}

That call uses the fast profile. It returns container metadata, streams, chapters, and embedded subtitles. It does not detect scenes, extract frames, or run speech recognition.

Then ask:

“Which chapter covers the pricing change?”

Cue returns chapter and subtitle locators with timestamp_ms and the source hash. It does not invent a transcript. A quote search matches embedded subtitles only. Render or crop a frame afterwards with video_evidence, once you have a timestamp.

Jobs Cue is built for

Ask your agent	Cue returns
“Find this quote.”	a timestamped match in embedded subtitles, when those subtitles exist
“Summarize this meeting.”	chapters and embedded subtitles
“Where is the code shown?”	one frame from `video_evidence` at a known timestamp
“What changed in this demo?”	scene boundaries when `profile` is `quality` or `include_scenes` is set
“Give me the useful moments.”	chapters, embedded subtitles, warnings, and gaps

Tool surface

Tool	Purpose
`read_video`	Default `fast` profile: container metadata, streams, chapters, and embedded subtitles. Scenes are opt-in.
`search_video`	Search embedded subtitle cues. It does not run speech recognition.
`video_evidence`	Named follow-up: render, crop, or OCR one frame at a timestamp

Predictable defaults

Omitted profile is fast: container metadata, streams, chapters, and embedded subtitles. No scenes, frames, or speech recognition.
profile quality adds ffmpeg scene detection. It does not extract frames or run speech recognition.
include_keyframes and include_transcript stay off. Setting either one on the shipped server returns a warning and an empty array.
OCR stays on video_evidence.
No cloud video API or frame-by-frame vision model is required.
Missing ffprobe or embedded subtitles is reported as a gap, not guessed around.

Why agents trust it

Every claim can point back to timestamp_ms, a stream index, a subtitle range, or the source hash. A frame index is present only after video_evidence.

Companion MCP tools

Product	Job
Citra	PDF answers with page-level proof
Iris	Image facts and pixel evidence
Spine	Repository architecture and impact
Locus	Exact code-chunk retrieval
Lookout	Web research with source excerpts

Each product is independent. Install only the tools your agent needs.

Development

bun install
bun run build
bun test
cargo test
bun run benchmark:public-proof
bun run benchmark:release-gate

License

MIT

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

Read the getting-started guide

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.

Cue