mcp-video-analyzer
mcp-video-analyzer extracts transcripts, key frames, OCR text, and metadata from video URLs (Loom, .mp4, .webm, .mov, etc.) with the following capabilities:
Full video analysis (
analyze_video): Extract everything at once — timestamped transcript with speaker IDs, deduplicated key frames via scene-change detection, OCR text from frames (code, UI text, error messages), an annotated timeline merging transcript + frames + OCR, metadata (title, duration, platform), viewer comments, chapters, and AI summary. Supportsbrief,standard, anddetaileddepth levels.Transcript extraction (
get_transcript): Pull only the timestamped transcript with speaker IDs; falls back to Whisper when no native transcript is available.Metadata retrieval (
get_metadata): Fetch metadata, comments, chapters, and AI summary without downloading the video or extracting frames.Frame extraction (
get_frames): Extract key frames via scene-change detection (default) or dense sampling (1 frame/sec), with deduplication and JPEG optimization.Single frame (
get_frame_at): Capture one frame at a specific timestamp to inspect exactly what's on screen.Burst frame extraction (
get_frame_burst): Extract N frames evenly across a narrow time range — ideal for analyzing motion, animations, or fast scrolling.Moment deep-dive (
analyze_moment): Focused analysis on a specific time range combining burst frames, filtered transcript, OCR, and an annotated timeline.Caching: Results cached in memory for 10 minutes; use
forceRefreshto bypass.Flexible output: Filter returned fields, control frame count, adjust scene-change sensitivity, and optionally return frames as base64 inline.
Enables analysis of Loom videos by extracting transcripts, key frames, metadata, and comments to provide a unified timeline of visual and audio content.
No existing video MCP combines transcripts + visual frames + metadata in one tool. This one does — across Loom, the major yt-dlp platforms (YouTube/Vimeo/TikTok/Instagram/X/Twitch/Dailymotion/Facebook), direct video URLs, and local files.
Want a full pipeline, not just a tool? social-knowledge-base is built on top of this server — it downloads whole Instagram creator accounts (reels, stories, highlights), transcribes them, and turns the result into a searchable, RAG-queryable knowledge base with AI-generated notes. Use this MCP when you want per-video analysis inside an agent; use social-knowledge-base when you want to archive and query an entire account.
Installation
Prerequisites
Node.js 22.12+ — required to run the server via
npxyt-dlp — required for YouTube/Vimeo/TikTok/Instagram/X/Twitch/Dailymotion/Facebook URLs; optional for everything else (improves Loom download quality). Install with
pip install yt-dlpChrome/Chromium (optional) — fallback for frame extraction if yt-dlp is unavailable
Without yt-dlp or Chrome, direct URLs and local files still get frames — the bundled
ffmpeg-staticdoes the extraction, and Loom falls back to its own CDN download. Platform URLs (YouTube etc.) degrade to a clear "install yt-dlp" warning. Transcripts, metadata, and comments never require either.
There are three ways in: the /video plugin (Claude Code — slash command + MCP server auto-configured), a plain MCP server config (any MCP client), or the portable skill + CLI (Codex, Cursor, Copilot, and any agent with a shell — no MCP required).
Claude Code — /video plugin (recommended)
/plugin marketplace add guimatheus92/mcp-video-analyzer
/plugin install video@mcp-video-analyzerThis adds the /video slash command and auto-registers the MCP server — no claude mcp add needed:
/video https://youtu.be/jNQXAC9IVRw what happens at 0:10?
/video ~/Movies/screen-recording.mp4 when does the UI break?Other agents — Codex, Cursor, Copilot, Gemini CLI, …
npx skills add guimatheus92/mcp-video-analyzerInstalls the video skill (Agent Skills format) into every agent detected on your machine. Agents without the MCP server configured fall back to the bundled CLI automatically — zero configuration.
Claude Code (MCP only)
claude mcp add video-analyzer -- npx mcp-video-analyzer@latestThen restart Claude Code or start a new conversation.
VS Code / Cursor
Add to your MCP settings file:
VS Code:
File → Preferences → Settings → search "MCP"or edit~/.vscode/mcp.json/%APPDATA%\Code\User\mcp.json(Windows)Cursor:
Settings → MCP Servers → Add
{
"servers": {
"mcp-video-analyzer": {
"type": "stdio",
"command": "npx",
"args": ["mcp-video-analyzer@latest"]
}
}
}Then reload the window (Ctrl+Shift+P → "Developer: Reload Window").
Claude Desktop
Add to your Claude Desktop config file:
macOS:
~/Library/Application Support/Claude/claude_desktop_config.jsonWindows:
%APPDATA%\Claude\claude_desktop_config.json
{
"mcpServers": {
"video-analyzer": {
"command": "npx",
"args": ["mcp-video-analyzer@latest"]
}
}
}Then restart Claude Desktop.
CLI (one-shot, no MCP client)
The same engine is exposed as a one-shot command — this is what the video skill uses on agents without MCP, and it works standalone in any terminal:
npx -y mcp-video-analyzer@latest analyze "https://youtu.be/jNQXAC9IVRw"stdout is a single JSON document — metadata, transcript, ocrResults, timeline, warnings, frameCount, and frames as { time, filePath, mimeType } entries pointing at JPEG key frames copied to --out (default: the per-user cache dir — %LOCALAPPDATA% on Windows, ~/Library/Caches on macOS, $XDG_CACHE_HOME or ~/.cache on Linux — under mcp-video-analyzer/<url-hash>/; set MCP_CACHE_DIR to an absolute path to relocate it). Unlike the temp dir this used to live in, nothing reaps that location, so frames persist until you delete them — the directories are created 0700. Progress streams on stderr, so stdout can be piped straight into a JSON parser. Partial failures land in warnings with exit code 0; only hard failures exit 1.
Flag | Description |
|
|
| Max key frames, 1–60 (default adapts to duration) |
| Width cap for emitted frames (default |
| Output filter — comma-separated subset: |
| Bypass the cache and re-analyze |
| Tesseract languages (default |
| Whisper overrides for the transcription fallback |
| Where frame images are copied |
Run with no arguments (npx mcp-video-analyzer@latest) to start the MCP stdio server — the CLI is purely additive.
Verify it works
Once installed, ask your AI assistant:
Analyze this video: https://www.youtube.com/watch?v=jNQXAC9IVRw(also works with an Instagram/TikTok/Loom link, a direct .mp4 URL, or a local file path). If the server is connected, it will automatically call the analyze_video tool.
Related MCP server: Loom Transcript MCP Server
Tools
Eight tools — the AI picks the cheapest one for the job and calls it automatically. Click any tool to expand its parameters and examples.
Tool | What it does |
| Full analysis: transcript + key frames + OCR + timeline + metadata |
| Batch version, one structured result per source (resumable) |
| Transcript only (native captions or Whisper fallback) |
| Metadata + comments + chapters, no download |
| Key frames only (scene-change or dense 1 fps) |
| Deep-dive on a time range (burst frames + transcript + OCR) |
| Single frame at a timestamp |
| N frames across a narrow window (motion/animation) |
Extracts everything from a video URL in one call:
> Analyze this video: https://www.youtube.com/watch?v=abc123...Returns:
Transcript with timestamps and speakers
Key frames extracted via scene-change detection (automatically deduplicated). For static clips with no scene cuts — e.g. talking-head Reels/Stories where only an on-screen text overlay changes — it automatically falls back to uniform temporal sampling so you still get frames (and OCR) instead of an empty result.
OCR text extracted from frames (code, error messages, UI text, prices/dates/CTAs visible on screen)
Annotated timeline merging transcript + frames + OCR into a unified "what happened when" view
Metadata (title, duration, platform)
Comments from viewers
Chapters and AI summary (when available)
The AI will automatically call this tool when it sees a video URL — no need to ask.
Options:
detail— analysis depth:"brief"(metadata + truncated transcript, no frames),"standard"(default),"detailed"(dense sampling, more frames)fields— array of specific fields to return, e.g.["metadata", "transcript"]. Available:metadata,transcript,frames,comments,chapters,ocrResults,timeline,aiSummarymaxFrames(1-60) — cap on extracted frames. Default scales with video duration atstandarddetail (~12 for ≤30s up to 60 for >10min); fixed 60 atdetailed, 0 atbrief. An explicit value always winsthreshold(0.0-1.0, default 0.1) — scene-change sensitivityforceRefresh— bypass cache and re-analyzeskipFrames— skip frame extraction for transcript-only analysismodel/language/initialPrompt— per-call Whisper overrides for the transcription fallback (overrideWHISPER_MODEL/WHISPER_LANGUAGE/WHISPER_PROMPTfor this call only — pick a heavier model or a domain glossary for one hard clip without restarting the server)
> Analyze every .mp4 in this folderRuns analyze_video over a list of sources with a concurrency limit (default 2), returning one structured result per source — counts + warnings on success, or a per-item error on failure (one bad file never aborts the batch). Frame images are not inlined and full transcript/OCR/timeline are returned only when fields is set; otherwise you get counts. Pair with MCP_WRITE_SIDECARS=1 (below) so each video's result persists to disk and a re-run resumes instead of recomputing.
> Get the transcript from this videoQuick transcript extraction. Falls back to Whisper transcription when no native transcript is available. Accepts the same per-call model / language / initialPrompt overrides as analyze_video.
> What's this video about?Returns metadata, comments, chapters, and AI summary without downloading the video.
> Extract frames from this video with dense samplingTwo modes:
Scene-change detection (default) — captures visual transitions
Dense sampling (
dense: true) — 1 frame/sec for full coverage
> Analyze what happens between 1:30 and 2:00 in this videoCombines burst frame extraction + filtered transcript + OCR + annotated timeline for a focused segment. Use when you need to understand exactly what happens at a specific moment.
> Show me the frame at 1:23 in this videoThe AI reads the transcript, spots a critical moment, and requests the exact frame to see what's on screen.
> Show me 10 frames between 0:15 and 0:17 of this videoFor motion, vibration, animations, or fast scrolling — burst mode captures N frames in a narrow window so the AI can see frame-by-frame changes.
Detail Levels
Level | Frames | Transcript | OCR | Timeline | Use case |
| None | First 10 entries | No | No | Quick check — what's this video about? |
| Duration-adaptive: ~12 (≤30s) up to 60 (>10min), scene-change | Full | Yes | Yes | Default — full analysis |
| Up to 60 (1fps dense) | Full | Yes | Yes | Deep analysis — every second captured |
Caching
Results are cached in memory for 10 minutes. Subsequent calls with the same URL and options return instantly. Use forceRefresh: true to bypass the cache. skipFrames is part of the cache and sidecar key, so a transcript-only analysis and a framed one of the same URL never answer for each other.
Persistent sidecars (resumable bulk processing)
The in-memory cache is lost on restart, which makes reprocessing a large local corpus costly. Set MCP_WRITE_SIDECARS=1 to also persist results next to each local video so the work survives restarts and can resume:
<stem>.vtt— the transcript, only when it was generated by the Whisper fallback (an existing<stem>.vttfrom your own pipeline is never overwritten). A later call reuses it via the normal sidecar reader and skips Whisper entirely.<stem>.analysis.json+<stem>.frames/— the full result (frames + OCR + timeline), keyed by the video'smtime:sizeand the analysis params. On a later call with a matching stamp + params, the result is returned straight from disk (no extraction, no OCR).
This makes analyze_videos over thousands of files resumable, and lets an external GPU transcription pipeline and this MCP share results through the filesystem: the pipeline writes <stem>.vtt, and the MCP picks it up instead of running Whisper.
Supported Sources
Source | Transcript | Metadata | Comments | Frames | Auth |
Loom | Yes | Yes | Yes | Yes (usually needs yt-dlp — see note) | None |
YouTube / Vimeo / TikTok / Instagram / X / Twitch / Dailymotion / Facebook | Native captions (uploaded > auto-generated) or Whisper fallback | Yes (title, duration, uploader, views, chapters, upload date) | No | Yes (capped at 1080p) | yt-dlp installed; cookies for Instagram / age-restricted (see below) |
Direct URL (.mp4, .mov, .mkv, .webm, …) | No | Duration only | No | Yes | None |
Direct URL + TwelveLabs | Yes (Pegasus, best-effort) | Duration floor + title | No | Yes |
|
Local file (absolute path or | Sidecar | Probed via ffmpeg (duration, dims, codec, audio presence) | No | Yes | None |
Loom frames: transcript, metadata, and comments come straight from Loom's API with no extra tooling. Frame extraction is different — Loom serves most videos as separate DASH video+audio streams, which only yt-dlp (
pip install yt-dlp) fetches and merges. Merging uses the bundledffmpeg-static, so no system ffmpeg is required. Without yt-dlp a direct-CDN fallback still covers some videos; when it can't, you get transcript + metadata + comments plus a warning explaining why frames are missing.Local files: pass an absolute path (e.g.,
/Users/you/clip.mp4) or afile://URI as theurlargument to any tool. Relative paths are rejected — the server's working directory is unpredictable from the MCP client. Note that any caller of the MCP server can ask it to read any file the server process has access to. UNC / network share paths (\\host\share\clip.mp4) are the exception: they reach the network rather than local disk, so they follow the network destination rules and needMCP_ALLOW_PRIVATE_URLS=1.Sidecar transcripts: if a
clip.vtt,clip.srt,clip.en.vtt, etc. lives next toclip.mp4, it's used as the transcript automatically — no Whisper roundtrip needed. SRT is converted to VTT in-memory.Embedded subtitles: if no sidecar is found and the container has an embedded subtitle stream (common in
.mkv/.mov/.mp4from screen recorders), it's transmuxed to VTT via ffmpeg and used as the transcript.Recognized extensions (local files and direct URLs):
.mp4.mov.mkv.webm.avi.m4v.wmv.flv.mpeg.mpg.m2ts.mts.3gp.ogv. The extension only gates routing — ffmpeg does the actual demuxing, so most common containers work..tsis excluded to avoid colliding with TypeScript source files.
Network destinations
Only http:// and https:// URLs are fetched, and only to public addresses. Requests to loopback (localhost, 127.0.0.1, ::1), private/LAN ranges, link-local, CGNAT, .local mDNS names, and Windows UNC paths are refused, as are non-HTTP schemes like ftp:// and data:.
This matters because the url argument is attacker-reachable in the normal case: an agent driving this server can be steered by the content it reads, so a URL it passes in is not necessarily one the user chose. Without the restriction the server is a proxy into whatever network it happens to sit on.
The check runs on the resolved address, not just the text, so a public hostname that resolves to 10.0.0.5 is refused too — and every hop of a redirect chain is re-checked, since a public URL answering 302 Location: http://127.0.0.1/ would otherwise walk straight past a first-hop-only check.
# Serving videos from a NAS or a local dev server? Opt back in:
MCP_ALLOW_PRIVATE_URLS=1Cloud instance metadata endpoints (169.254.169.254, Azure's 168.63.129.16, and friends) stay blocked even with that set — there is no legitimate video there, and they are what an SSRF is usually after.
Known limitation: the address is checked at resolution time, not at connection time, so DNS rebinding — a domain answering a public address to the check and a private one to the connection — is not covered. Run the server behind an egress proxy if that is in your threat model.
Platform URLs via yt-dlp (YouTube, Instagram, TikTok, …)
Single-video pages on major platforms route through yt-dlp (pip install yt-dlp — required for these URLs). Playlists, channels, and profile pages are rejected by design; pass individual video URLs (batch them with analyze_videos).
Transcript: native captions are preferred and free — uploaded subtitles first, auto-generated captions as fallback (rolling-window duplication is collapsed).
WHISPER_LANGUAGE(e.g.pt) is also used to pick the caption language. Videos with no captions at all fall through to the normal Whisper chain.Metadata: title, duration, uploader/channel, view count, upload date, and chapters — no download needed.
Download: capped at 1080p (frames/OCR don't need more), live streams are skipped, and DASH audio+video is merged with the bundled
ffmpeg-static(no system ffmpeg required).Cookies — Instagram and age-restricted videos usually require a logged-in session:
Env var | What it does | Example |
| Cookie file (Netscape format), wins when both are set |
|
| Extract cookies from an installed browser |
|
Browser cookie extraction requires the browser to be closed on Windows (the cookie database is locked while it runs). If that's inconvenient, export a
cookies.txtonce (e.g. with a "Get cookies.txt" browser extension) and pointYTDLP_COOKIESat it. Private/age-restricted videos without valid cookies don't crash the tool — the yt-dlpERROR:line surfaces inwarnings[].
TwelveLabs Pegasus (optional)
Set the TWELVELABS_API_KEY environment variable to analyze direct video URLs with TwelveLabs Pegasus. Pegasus analyzes the video server-side (visuals and its own audio) and returns an AI-generated, timestamped transcript plus an AI summary as text — capabilities the DirectAdapter can't provide (a raw .mp4 URL has no transcript or summary on its own), and with no Whisper key required.
The transcript is best-effort LLM output, not a deterministic ASR dump: Pegasus is prompted to emit [MM:SS] line rows, and lines that don't match that shape are dropped, so wording and exact timestamps depend on the model's prompt adherence. Failures (bad key, timeout, API error) surface in the tool's warnings[] rather than silently returning an empty transcript.
The biggest win is on the text-only paths: get_transcript and get_metadata return a Pegasus transcript and summary for direct URLs — a few KB of text, no frame images, no per-frame token cost. analyze_video at detail: "standard"/"detailed" still extracts frames in addition (use detail: "brief" to stay text-only).
Long videos: the summary and full transcript share a single capped completion (
max_tokens= 16384), so for very long videos the transcript may be truncated. For multi-hour content, chunking by time window is the better approach.
It's fully opt-in and non-breaking: when TWELVELABS_API_KEY is set the TwelveLabsAdapter handles direct video URLs (it registers the public URL with TwelveLabs — no upload); when it's unset, the DirectAdapter handles them exactly as before. Loom URLs are unaffected. Get a key at playground.twelvelabs.io.
Transcription (Whisper fallback)
When a source has no native transcript (no sidecar .vtt/.srt, no embedded subtitles, no platform captions), the audio track is transcribed with Whisper via a graceful fallback chain (in execution order):
Silent tracks: before any Whisper run, the audio is probed with ffmpeg
volumedetect(first 2 minutes). A present-but-mute track — common in muted Reels/Stories — skips transcription entirely and emits a warning that the empty transcript is expected content, not an error, saving a pointless Whisper run.
@huggingface/transformers (JS-native, zero external deps) — opt-in only: this strategy runs first, but only when
WHISPER_HF_MODELis explicitly set. When it's unset (the default) the strategy is skipped entirely, so the CLI below wins and itsWHISPER_MODEL/WHISPER_LANGUAGEsettings are never silently overridden.whisperCLI — used when awhisperexecutable is found (pip install -U openai-whisper). PointWHISPER_BINat the executable if it isn't onPATH. Model viaWHISPER_MODEL, language viaWHISPER_LANGUAGE. The bundledffmpeg-staticis put on the CLI'sPATHautomatically, so no system ffmpeg is required.OpenAI Whisper API — used when
OPENAI_API_KEYis set.
No backend configured? If none of the three is available (no
whisperonPATH/WHISPER_BIN, noOPENAI_API_KEY, noWHISPER_HF_MODEL), transcription tools return an empty transcript with a warning telling you how to enable one — rather than a silent "no transcript". Installopenai-whisperor set one of the keys above. (The CLI is spawned withPYTHONUTF8=1so non-English/CJK transcripts don't crash the Python process on Windows.)
Env var | Applies to | Default | Example |
|
|
|
|
|
| auto-detect |
|
|
| — |
|
|
|
|
|
|
| — |
|
|
| — |
|
|
| — |
|
|
| off |
|
| HF transformers (opt-in) | — (strategy off) |
|
| OpenAI API | — |
|
The default
tinymodel is fast but weak for non-English audio. For Portuguese (or other non-English) sources, install the CLI and setWHISPER_MODEL=small(ormedium) +WHISPER_LANGUAGE=ptfor much better accuracy. AddWHISPER_PROMPTwith a domain glossary (brand/place names) to fix proper nouns. You can also overridemodel/language/initialPromptper call onanalyze_video/get_transcript/analyze_videos— no restart needed.GPU (faster-whisper):
whisper-ctranslate2(pip install -U whisper-ctranslate2) is a drop-in CLI with the same flags plus--device cuda/--compute_type/--beam_size. PointWHISPER_BINat it and setWHISPER_DEVICE=cuda(+ optionallyWHISPER_COMPUTE=float16). These GPU flags are env-gated — they're only passed when set, so plainopenai-whisper(which rejects--compute_type) keeps working when they're unset.Windows note: pip installs
whisper.exeinto the PythonScripts/dir, which is often not on thePATHthat GUI-launched MCP clients inherit. If transcripts come back empty, setWHISPER_BINto the full path ofwhisper.exe.
Frame Extraction Strategies
Frame extraction uses a two-strategy fallback chain — no single dependency is required:
Strategy | How it works | Speed | Requirements |
yt-dlp + ffmpeg (primary) | Downloads video, extracts frames via scene detection | Fast, precise | yt-dlp ( |
Browser (fallback) | Opens video in headless Chrome, seeks to timestamps, takes screenshots | Slower, no download needed | Chrome or Chromium installed |
The fallback is automatic — if yt-dlp is not available, the server tries browser-based extraction via puppeteer-core. If neither is available, analysis still returns transcript + metadata + comments, just no frames.
Post-Processing Pipeline
After frame extraction, the pipeline automatically applies:
Step | What it does | Why |
Frame deduplication | Removes near-identical consecutive frames using perceptual hashing (dHash + Hamming distance) | Screencasts often have long static moments — dedup removes redundant frames, saving tokens |
OCR | Extracts text visible on screen from each frame (via tesseract.js). Each frame is first preprocessed — grayscale + 2× upscale + contrast normalization + sharpen — which materially improves accuracy on stylized overlays (prices, dates, coupons, CTAs). | Captures code, error messages, terminal output, UI text that the transcript doesn't cover |
Annotated timeline | Merges transcript timestamps + frame timestamps + OCR text into a single chronological view | Gives the AI a unified "what was said, what changed visually, and what text appeared" at each moment |
The OCR step requires tesseract.js (included as a dependency). If it fails to load, analysis continues without OCR — no frames or transcript are lost. OCR preprocessing is on by default; set MCP_OCR_PREPROCESS=0 to OCR the raw frames instead.
OCR always reads the full-resolution frame, not the copy emitted to the client. The two have different jobs: the emitted frame is capped for token cost, while recognition needs every pixel it can get.
Frame size (dense UI captures)
Emitted frames are capped at 800 px wide, which suits the common case — talking-head clips, Reels, bug repros — where the subject fills the frame.
It is the wrong size for a dense UI capture: a terminal, dashboard, IDE or spreadsheet recording, where the meaning lives in small text. An unscaled 1920×1080 screen recording lands at 800×450, and a 15 px UI font drops below what a vision model can resolve.
Pass maxWidth per call to keep more (or all) of the source resolution — 0 disables the cap:
get_frames(url, { maxFrames: 8, maxWidth: 0 }) // source resolution
get_frame_at(url, "2:14", { maxWidth: 1920 })
analyze_video(url, { detail: "standard", maxWidth: 1568 })Supported on analyze_video, analyze_videos, analyze_moment, get_frames, get_frame_at and get_frame_burst, and on the CLI as --max-width <px>.
Native frames cost several times more context than the default, so raise the cap deliberately — get_frames returns up to 20 frames and analyze_video at detailed up to 60.
Variable | Applies to | Default | Notes |
| Emitted frame width, in px |
|
|
| Emitted frame JPEG quality |
| Raise it when thin glyphs matter; env only, there is no per-call quality parameter. Values outside 1–100 fall back |
| Root for the tessdata cache and the CLI's default | per-user cache dir | Absolute paths only (a relative value is ignored). Use it when |
| Reaching private/loopback network addresses | off |
|
A value either variable can't use — 1e3, 1920px, a quality of 150 — is rejected with a one-time warning on stderr and the default applies. It is not silently accepted: the whole point of the setting is to escape a downscale that otherwise looks like a normal result.
Prefer the per-call parameter: the server starts once per session, so an environment variable cannot differ between an overview of a YouTube clip and a close read of a screen recording. The width a call actually uses is part of the cache and sidecar key, so analyzing the same video at 800 px and then at maxWidth: 0 re-runs the pipeline instead of returning the first result twice.
Complementary Tools
Chrome DevTools MCP
For live web debugging alongside video analysis, pair this server with the Chrome DevTools MCP:
claude mcp add chrome-devtools npx @anthropic-ai/mcp-devtools@latestWhen to use each:
Scenario | Tool |
Bug report recorded as a Loom video |
|
Live debugging a web page | Chrome DevTools MCP — inspect DOM, console, network, take screenshots |
Video shows UI issue, need to reproduce it | Use both: analyze the video first, then open the page in Chrome DevTools to reproduce |
The two MCPs complement each other: video analyzer understands recorded content, DevTools interacts with live pages.
Example Output
The examples/loom-demo/ folder contains real outputs from analyzing a public Loom video (Boost In-App Demo Video, 2:55).
File | What it shows |
Title, duration, platform | |
42 timestamped entries with speaker IDs | |
Unified chronological view (transcript + frames merged) | |
Filtered transcript for | |
Complete |
Frame images (19 total in examples/loom-demo/frames/):
scene_*.jpg— scene-change detection (key visual transitions)dense_*.jpg— 1fps dense sampling (every 10th frame saved as sample)burst_*.jpg— burst extraction for moment analysis (0:30–0:45)
Regenerate after changes:
npx tsx examples/generate.ts— requires yt-dlp + network access.
Development
# Install dependencies
npm install
# Run all checks (format, lint, typecheck, knip, tests)
npm run check
# Audit dependencies. `security` covers what the published package ships and
# is a blocking CI job; `security:all` adds devDependencies. Both also run on
# a weekly cron, because npm audit reads a live advisory database.
npm run security
# Build
npm run build
# Run E2E tests (requires network; add WHISPER_E2E=1 to include the
# transcription outcome test — needs a whisper CLI installed)
npm run test:e2e
# Just the video-format matrix: a real clip per container/codec
# (mp4/h264+hevc+av1, webm, mkv, mov, avi, m4v, mpeg, mpg, m2ts, mts,
# 3gp, ogv, flv, wmv) decoded end to end. ~15s on a warm cache; the
# first run fetches ~7MB of tesseract traineddata.
npm run test:formats
# Build + boot the real MCP server/CLI (seconds)
npm run test:smoke
# Everything: check → e2e → smoke → verify-package
npm run verify-all
# Open MCP Inspector for manual testing
npm run inspectArchitecture
src/
├── index.ts # Entry point (shebang + stdio)
├── server.ts # FastMCP server + tool registration
├── tools/ # MCP tool definitions (7 tools)
│ ├── analyze-video.ts # Full analysis with detail levels + caching
│ ├── analyze-moment.ts # Deep-dive on a time range
│ ├── get-transcript.ts # Transcript-only with Whisper fallback
│ ├── get-metadata.ts # Metadata + comments + chapters
│ ├── get-frames.ts # Frames-only (scene-change or dense)
│ ├── get-frame-at.ts # Single frame at timestamp
│ └── get-frame-burst.ts # N frames in a time range
├── adapters/ # Source-specific logic
│ ├── adapter.interface.ts # IVideoAdapter interface + registry
│ ├── loom.adapter.ts # Loom: authless GraphQL
│ ├── local-file.adapter.ts # Local files: absolute path or file:// URI
│ ├── twelvelabs.adapter.ts # TwelveLabs Pegasus: transcript + AI summary (opt-in)
│ └── direct.adapter.ts # Direct URL: any mp4/webm link
├── processors/ # Shared processing
│ ├── frame-extractor.ts # ffmpeg scene detection + dense + burst extraction
│ ├── browser-frame-extractor.ts # Headless Chrome fallback for frames
│ ├── audio-transcriber.ts # Whisper fallback (HF transformers → CLI → OpenAI)
│ ├── image-optimizer.ts # sharp resize/compress
│ ├── frame-dedup.ts # Perceptual dedup (dHash + Hamming distance)
│ ├── frame-ocr.ts # OCR text extraction (tesseract.js)
│ └── annotated-timeline.ts # Unified timeline (transcript + frames + OCR)
├── config/
│ └── detail-levels.ts # brief / standard / detailed config
├── utils/
│ ├── cache.ts # In-memory TTL cache with LRU eviction
│ ├── field-filter.ts # Selective field filtering for responses
│ ├── url-detector.ts # Platform detection from URL
│ ├── vtt-parser.ts # WebVTT → transcript entries
│ └── temp-files.ts # Temp directory management
└── types.ts # Shared TypeScript interfacesBurst timing
Burst extraction accepts fractional timestamps (for example, 0:01.1). The
FFmpeg path selects up to the requested count of existing source frames within
[from, to), without duplicating frames to satisfy an excessive sampling rate.
Processor results retain each selected frame's video-relative presentation time,
including fractional seconds. Browser fallback preserves fractional seek targets
and separate screenshot files; its times are requested seeks, not measured source
frame PTS. Browser seeking is not a frame-accurate substitute for local decoding.
get_frame_burst returns a JSON text block containing
{ frameCount, from, to, warnings, frames: [{ time, timingOrigin }] }, followed
by images in exactly the same order as frames. timingOrigin: "source-pts"
means measured video-relative FFmpeg presentation time; "seek-target" means
requested browser seek time. Empty/degraded results contain frames: [].
License
MIT
Available Tools
8 toolsanalyze_momentARead-onlyIdempotent
Deep-dive analysis of a specific time range in a video.
Combines burst frame extraction + transcript filtering + OCR + annotated timeline for a focused segment of the video.
Use this when you need to understand exactly what happens between two timestamps:
What's on screen (frames + OCR text extraction)
What's being said (transcript filtered to the range)
Unified timeline merging visual and audio content
Example: analyze_moment(url, "1:30", "2:00", 10) → 10 frames + transcript + OCR for that 30s window
Supports: Loom (loom.com/share/...), YouTube/Vimeo/TikTok/Instagram/X/Twitch/Dailymotion/Facebook (requires yt-dlp), direct video URLs (.mp4, .webm, .mov), and local video files (absolute path or file:// URI).
Inputs: to (required), url (required), from (required), count, +2 more.
analyze_videoARead-onlyIdempotent
Analyze a video URL to extract transcript, key frames, metadata, comments, OCR text, and annotated timeline.
Returns structured data about the video content:
Transcript with timestamps and speakers
Key frames extracted via scene-change detection (deduplicated, as images). For static clips with no scene cuts (e.g. talking-head Reels/Stories where only on-screen text changes) it automatically falls back to uniform temporal sampling.
OCR text extracted from frames (code, error messages, UI text, prices/dates/CTAs visible on screen)
Annotated timeline merging transcript + frames + OCR into a unified chronological view
Metadata (title, duration, platform)
Comments from viewers (if available)
Supports: Loom (loom.com/share/...), YouTube/Vimeo/TikTok/Instagram/X/Twitch/Dailymotion/Facebook (requires yt-dlp), direct video URLs (.mp4, .webm, .mov), and local video files (absolute path or file:// URI).
Detail levels:
"brief": metadata + truncated transcript only (fast, no video download)
"standard": full analysis with scene-change frames (default)
"detailed": dense sampling (1 frame/sec), more frames, full OCR
Use options.fields to request only specific data (e.g., ["metadata", "transcript"]). Use options.forceRefresh to bypass the cache. Use options.model / options.language / options.initialPrompt to override Whisper transcription per call (e.g. a heavier model + a domain glossary for hard audio) without restarting the server.
Inputs: url (required), options.
analyze_videosARead-onlyIdempotent
Batch-analyze many videos in one call, with a concurrency limit and per-item results.
For each source it runs the same pipeline as analyze_video (frames + OCR + transcript + timeline), reusing the shared cache and on-disk sidecars. Designed for processing a corpus of local files: pair it with MCP_WRITE_SIDECARS=1 so results persist next to each video and a re-run resumes instead of recomputing.
Returns a JSON summary plus one structured entry per source:
ok=true → title, duration, frameCount, ocrCount, transcriptEntries, warnings
ok=false → the error message for that specific video (other videos still complete)
To keep the response bounded, frame images are NOT inlined and full transcript/OCR/timeline arrays are returned only when options.fields is set; otherwise you get counts. Use analyze_video on an individual source when you need the images or full data inline.
Inputs: sources (required), options, concurrency.
get_frame_atARead-onlyIdempotent
Extract a single video frame at a specific timestamp.
Useful for inspecting what's on screen at a particular moment. The AI reads the transcript, identifies a critical moment, and requests the exact frame at that timestamp.
Supports: Loom (loom.com/share/...), YouTube/Vimeo/TikTok/Instagram/X/Twitch/Dailymotion/Facebook (requires yt-dlp), direct video URLs (.mp4, .webm, .mov), and local video files (absolute path or file:// URI).
Args:
url: Video source (URL or local path)
timestamp: Time position (e.g., "1:23", "0:05", "01:23:45")
Returns: A single image of the video frame at the specified timestamp.
Inputs: url (required), timestamp (required), maxWidth, returnBase64.
get_frame_burstARead-onlyIdempotent
Extract up to the requested number of source frames across a time range.
Designed for motion and vibration analysis where scene-change detection fails because the "scene" doesn't change — only the position/state of objects does.
Example: get_frame_burst(url, "0:15", "0:17", 10) → 10 frames in 2 seconds
AI sees the object in different positions across frames → understands the vibration
Works for: shaking, flickering, animations, fast scrolling, loading spinners
Supports: Loom (loom.com/share/...), YouTube/Vimeo/TikTok/Instagram/X/Twitch/Dailymotion/Facebook (requires yt-dlp), direct video URLs (.mp4, .webm, .mov), and local video files (absolute path or file:// URI).
Args:
url: Video source (URL or local path)
from: Start timestamp (e.g., "0:15")
to: End timestamp (e.g., "0:17")
count: Number of frames (default: 5, max: 30)
Returns: A JSON text block with { frameCount, from, to, warnings, frames: [{ time, timingOrigin }] }, followed by images in the same order as frames. timingOrigin is "source-pts" for measured video-relative FFmpeg timestamps and "seek-target" for requested browser seeks. Empty results include frames: []. Up to N source-frame images sampled in [from, to) by ffmpeg, without duplicating frames to meet the requested count. Fractional bounds such as "0:01.1" are supported. Browser fallback seeks across the range, including its endpoints; those seek targets are not measured source-frame PTS.
Inputs: to (required), url (required), from (required), count, +2 more.
get_framesARead-onlyIdempotent
Extract key frames from a video URL without transcript or metadata.
Two extraction modes:
Scene-change detection (default): captures visual transitions
Dense sampling (dense=true): captures 1 frame/sec for full video coverage
Returns optimized, deduplicated JPEG frames.
Supports: Loom (loom.com/share/...), YouTube/Vimeo/TikTok/Instagram/X/Twitch/Dailymotion/Facebook (requires yt-dlp), direct video URLs (.mp4, .webm, .mov), and local video files (absolute path or file:// URI).
Inputs: url (required), options.
get_metadataARead-onlyIdempotent
Get video metadata, comments, chapters, and AI summary from a video URL.
Returns structured metadata without downloading the video or extracting frames. Faster than analyze_video when you only need metadata.
Supports: Loom (loom.com/share/...), YouTube/Vimeo/TikTok/Instagram/X/Twitch/Dailymotion/Facebook (requires yt-dlp), direct video URLs (.mp4, .webm, .mov), and local video files (absolute path or file:// URI).
Inputs: url (required).
get_transcriptARead-onlyIdempotent
Extract only the transcript from a video URL.
Returns timestamped transcript entries with speaker identification (when available). Faster than analyze_video when you only need the transcript.
If the platform has no native transcript, attempts Whisper fallback transcription (requires @huggingface/transformers, whisper CLI, or OPENAI_API_KEY).
Supports: Loom (loom.com/share/...), YouTube/Vimeo/TikTok/Instagram/X/Twitch/Dailymotion/Facebook (requires yt-dlp; native captions preferred), direct video URLs (.mp4, .webm, .mov), and local video files (absolute path or file:// URI). For local files a sidecar .vtt/.srt next to the file is used first, then an embedded subtitle track, and only then the Whisper fallback if neither exists.
Inputs: url (required), options.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 8 tools
Each tool has a mostly distinct purpose: analyze_video (full pipeline), get_metadata, get_transcript, get_frames, get_frame_at, get_frame_burst, analyze_moment, and analyze_videos. The main risk is overlap between analyze_moment and a manual combination of get_frame_burst + get_transcript, and between get_frames and get_frame_burst, but the descriptions clearly differentiate sampling modes and use cases.
All names use consistent snake_case with two coherent verb families: analyze_* for heavy processing and get_* for targeted extraction. The only slight wrinkle is analyze_video vs analyze_videos (singular/plural batch pair), which is a conventional but potentially confusing distinction.
8 tools is well-scoped for a video analysis server. Each tool earns its place by covering a distinct granularity (whole video, moment, single frame, frame burst, key frames, metadata, transcript, batch).
The surface covers the domain thoroughly: full analysis, metadata, transcript, multiple frame extraction modes, focused segment analysis, and batch processing. Minor gaps like a standalone OCR-only or audio-only extraction exist, but the main video-analysis workflows have no dead ends.
Maintenance
Related MCP Connectors
Multimodal video analysis MCP — transcription, vision, and OCR for any video URL.
Real-time video transcripts, RAG captions, metadata, search & comments MCP. OAuth or API key.
Any social-video URL → transcript, metadata, frames, OCR, summary, search, Q&A. MCP server + x402.
MCP server for Clipkit — gives AI agents a video toolbox via the Clipkit schema.
Related MCP Servers
- AlicenseBqualityDmaintenanceMCP (Model Context Protocol) server that utilizes the Google Gemini Vision API to interact with YouTube videos. It allows users to get descriptions, summaries, answers to questions, and extract key moments from YouTube videos.425 npm6MIT
- FlicenseNot gradedqualityNot gradedmaintenanceExtracts transcripts, titles, descriptions, and comments from Loom videos via a remote MCP server.-
- AlicenseAqualityDmaintenanceAn MCP server that enables AI agents to analyze videos locally by extracting transcripts, detecting scene changes, and returning key frames.57MIT
- AlicenseNot gradedqualityDmaintenanceAn MCP server for comprehensive video analysis — AI-powered transcription, visual frame analysis, and metadata extraction from 1000+ platforms.1MIT