
ClipIt: a Claude skill that turns long videos into short-form clips and clean, beat-synced fan edits
ClipIt: a Claude skill that turns long videos into short-form clips and clean, beat-synced fan edits (9:16)
·
An open-source clipping studio and a skill for Claude. It turns long videos into short-form vertical clips and fan edits:
crisp4k, hdr and bright.enhance / cine looks.study measures other edits (pacing, grade changes, loudness, drops) and makes sheets to read them by eye; match-look fits a colour LUT to a reference's grade; locate finds the raw moment a short was cut from.Everything runs locally with Python, OpenCV and ffmpeg. The tool has no telemetry and uploads nothing; its only network use is the one-time setup, the speech model (when a job needs it) and any links you ask it to fetch. It also understands effect names from popular desktop and mobile editors and rebuilds them (docs/TRANSLATE.md).
Inside Claude Code, run:
/plugin marketplace add shasshhaank/ClipIt
/plugin install clipit@clipitThat's it. The first time you ask for an edit, Claude asks once to set up the free tools it needs (a few minutes, about 700 MB, all in ~/.clipit). No Homebrew, no admin rights, nothing to run in a terminal. The speech model (about 1.5 GB) downloads only when a job needs transcription, and the optional subject cut-out (about 200 MB) only if you ask for text behind a person.
Download clipit-skill.zip from the releases, then in Claude go to Customize > Skills > + > Upload a skill, then type /clipit in a chat. Code execution must be on. In chat, Claude works inside a sandbox: upload your clips and song (it can't download from YouTube there), keep edits short (under about 20 s), and captions depend on whether the speech model can be downloaded. For long or high-resolution edits, use Claude Code.
git clone https://github.com/shasshhaank/ClipIt.git
mkdir -p ~/.claude/skills && ln -s "$PWD/ClipIt/skills/clipit" ~/.claude/skills/clipit
bash ClipIt/skills/clipit/install.shType /clipit and Claude starts: it asks for the footage, the music and any text, then edits. Or ask in your own words, something like "clip the most controversial 40 seconds of input/podcast.mp4 with captions" or "make a comeback edit of these videos on music/song.mp3: first the critics, then the century". If the request is vague, Claude decides for itself: it researches the story (and verifies every number it puts on screen), finds and downloads real motion footage, picks the song section, builds the edit, reviews contact sheets of its own render and tells you what it chose.
The skill's commands also work on their own:
PY=~/.clipit/venv/bin/python; CLIP=skills/clipit/clip.py
$PY $CLIP testmedia # synthetic footage + beat + placeholder logo
$PY $CLIP fxdemo # labelled reel of every effect -> output/fx_demo.mp4
$PY $CLIP story skills/clipit/examples/demo_story.json --relative
$PY $CLIP find "india vs west indies odi highlights" # research footage: candidates with size; no download
$PY $CLIP fetch https://youtu.be/VIDEO_ID --section 3:10-3:40 --res 1080
$PY $CLIP analyze input/podcast.mp4 --shots # transcript, ranked moments, hook lines, shots
$PY $CLIP sheet input/podcast.mp4 --n 30 # contact sheet for picking shots
$PY $CLIP talk input/podcast.mp4 --start 812 --end 851 --title "HE SAID WHAT?" --music music/bed.mp3
$PY $CLIP edit input/game.mp4 --music music/phonk.mp3 --length 14 --flow
$PY $CLIP plan-render work/podcast/plan.json output/v2.mp4 --draft # fast preview; every render also writes a contact sheet
$PY $CLIP study https://youtu.be/VIDEO_ID input/ref.mp4 # measure reference edits -> refs/<name>/report.md
$PY $CLIP match-look input/ref.mp4 input/game.mp4 --out looks/ref.cube # use the .cube as a "look"
$PY $CLIP locate input/full_stream.mp4 input/short.mp4 # where a short was cut fromOutputs go to ./output/ and intermediates to ./work/, both relative to where you run the command.
When a render is done, upload it to postxport.com to share it and send it for review before you post.
| Stage | Module | Approach |
|---|---|---|
| Transcribe | transcribe.py | Whisper large-v3-turbo with word timestamps |
| Moments | moments.py | polarizing lexicon and phrases, hook openers, speech pace and loudness peaks |
| Shots | moments.py | PySceneDetect cuts, a motion score per shot, and a photo-slideshow detector |
| Framing | reframe.py | YuNet face detection, active-speaker choice, per-scene zoom limits, dead-zone smoothing |
| Music | beats.py | librosa beats, downbeats from low-end phase, and the drop as the biggest energy jump |
| Plan | planner.py, story.py | cut grid, speed curves, flowing cuts, effect choreography and the audio mix |
| Render | render.py, fx.py, flow.py | frame-accurate time remap, one-warp camera, effects, grade, text and image layers, piped to ffmpeg |
| Overlays | captions.py, assets.py | Pillow text (no libass), cards, buttons, cursor and logo cut-outs |
| References | study.py | scene cuts, look and grade-change measures, loudness and drops; audio cross-correlation (locate); LUT fitting (match-look) |
Docs: PLAYBOOK.md (the editing rules), EFFECTS.md (every effect), STORY_SPEC.md (the story format), TRANSLATE.md (other editors' names mapped to ours).
fetch from a link. Copyrighted songs on fan pages are often claimed or muted.docs/TRANSLATE.md.Face detection uses OpenCV's YuNet model (MIT license, from the OpenCV Zoo). The display fonts the installer fetches (Anton, Archivo Black, Space Mono, Amatic SC, Montserrat, Noto Sans Devanagari) are under the SIL Open Font License.
MIT © Shashank Pandey · LinkedIn. See LICENSE.
This repo does not document an install step we could read. Check the README on GitHub.
Paste this prompt into your agent. It reads this page and does the setup for you.
Read https://aiagentslisting.com/mcp/clipit to learn what the "ClipIt" MCP server does and how to install it. Add it to my coding agent's MCP configuration as documented on that page, then confirm the server connects and list the tools it exposes.Agents can also browse this directory over MCP at https://aiagentslisting.com/api/mcp
This server doesn't publish a machine-readable tool manifest, and it doesn't expose a public endpoint we could ask. Connect the server locally to see what it exposes.
We could not run this server: it ships no npx package for our sandbox and no public endpoint. The four checks that need a live server are left out, and the grade tops out at B until one exists.
Maintain this server? Claiming proves you own the listing and earns the last 4 points.
Claim this listingPaste this line near the top of your repository's README. The badge always renders the current score.
[](https://aiagentslisting.com/mcp/clipit)Nothing comparable is listed yet.
v1.5ClipIt 1.5Release notesv1.4ClipIt 1.4Release notesv1.3ClipIt 1.3Release notesv1.2ClipIt 1.2Release notesv1.1ClipIt 1.1Release notesConnect to the AI Agents Listing MCP server at https://aiagentslisting.com/api/mcp using the HTTP transport and complete the sign-in when it asks. Then call the submit_review tool for the listing "clipit" with a rating from 1 to 5 and a short note on what worked and what did not.No reviews yet
Be the first to share your experience with this tool.
One email a week. New agents, MCP servers and skills, and what is actually getting traction.