Hand BaoCut to your agent (Claude Code, Codex, Cursor and others) and say what you want in one sentence: transcribe, subtitle, translate and dub, cut a talking-head recording, or make an explainer from scratch. The agent edits through commands, and the result is a timeline you can keep editing. Open the editor, watch it once, then tell the agent what to change or drag things yourself.
Quick start
1. Install BaoCut and open it once
Download for macOS (Apple Silicon) or Windows. On first launch the app puts the baocut command in your terminal. Check it:
baocut status
It lists what this machine can do right now and what is missing (local models, ffmpeg and so on), each with a fix.
If the terminal cannot find baocut (on Windows, open a new terminal first), install it once by full path (macOS shown); it tells you whether a folder still needs adding to PATH. For other cases see the desktop CLI entrypoint.
/Applications/BaoCut.app/Contents/MacOS/baocut-cli cli install
2. Install the BaoCut skill for your agent
baocut skill install
The skill is a manual on how to use BaoCut. Once it is installed, your agent knows when and how to use it. It goes into ~/.agents/skills and ~/.claude/skills by default and updates with the app; use --dir <folder> for another location. For agents that connect over MCP, run baocut mcp install.
3. Ask your agent
Transcribe ~/Movies/talk.mp4, add English and Chinese subtitles, and export an MP4.
Download the video at this link, translate it into English and dub it: <video page link>
Cut the filler words and long pauses from this recording, then make 3 vertical shorts.
Use BaoCut to make a 3-minute explainer about <topic>, presented by a virtual host.
4. Review and fine-tune
When the timeline plays from start to finish, the agent stops and gives you a review link. It opens this video's editor in your browser. Say what to change ("the subtitle at 0:42 covers the face"), or drag clips and fix captions yourself; you and the agent edit the same timeline. Once you approve, it exports the final video.
No external agent? That works too. Chat in the desktop app's Home. It runs on the Codex CLI or Claude Code signed in on this machine, and works the same way.
Why "built for agents"
Traditional editors are designed for human hands, so an agent has to work around the interface. Most AI video tools go the other way: they render one MP4, and if the second sentence of the translation is wrong, you start over. BaoCut lets the agent edit through commands and keeps the result on a timeline:
- Editing runs on commands, not clicks. The CLI, the MCP server and the skill share one tool catalog, and every command documents its parameters and effect. The agent never has to click through an interface.
- The result is an editable video. Transcripts, translations, subtitle layers, chapters, voice-overs and motion graphics all live inside one video. SRT, MP4, audio and transcript files are exports of it.
- Change one thing, and only that thing changes. Fix one subtitle line, swap one shot, redo one voice-over sentence. The agent edits the scope you named and returns a receipt, and a wrong edit can be undone. Render caches are invalidated by actual dependencies; a local edit sometimes still re-renders the compositions it touches.
- You and the agent share one editor. You drag a clip, the agent rewrites a caption; both go through the same command gateway, transactions and undo stack. You can take over at any point.
- Bounded autonomy. The agent works inside an explicit scope, access mode and budget, and destructive actions follow approval rules. Outside material such as a downloaded page or a transcript is treated as data only.
- Honest delivery. The agent reports what it measured, which frames it actually looked at, and what still needs a human ear. A failed check is reported as failed.
What it can do
From existing footage
- Transcribe with local models (Whisper large-v3 / v3-turbo, Qwen3-ASR, MOSS with diarization) or online APIs, with word-level timing. Segments stream into the timeline while the job runs.
- Subtitle in the original language, translated, or bilingual. The agent translates sentence by sentence itself; subtitle styles are remembered per project.
- Dub into another language. Translate, separate vocals from background (HTDemucs), synthesize per sentence with a cloned or preset voice, lay the new track on the timeline.
- Cut a talking-head recording. Remove filler words, long pauses, retakes and false starts on the timeline. The original media is never touched.
- Chapters, summaries, titles, blog posts, shorts. Read the timed transcript and write the derivatives, with every claim traceable to a timestamp.
- Fix and prepare footage on this computer. Stabilize shaky clips (vid.stab through your ffmpeg, or deshake), upscale video (Real-ESRGAN), make smooth slow motion (RIFE), grade shots to one reference, cut subjects out without a green screen (rembg's weights), and split any track into stems (HTDemucs). Each one writes a new file and leaves the original alone.
- Find the edit points. Detect shots, including handheld footage and fades, find dead air without a transcript, and read the beat, bars and accents of a song so cuts land on it.
From a brief
This is the beginning of the README. Read the rest at the original. Read the full README ↗
