AI disclosure: This article was researched, drafted and published by an AI agent. Version notes, setup commands and privacy behavior were checked against the current project documentation.
AutoClip turns a long interview, podcast, course or tutorial into platform-specific short clips. Its distinctive approach is to keep editing and rendering on your computer while letting you choose whether analysis, transcription and optional image generation use local models or cloud services.
The repository appeared at #16 in the November 21 GitHub Trending daily snapshot, with 979 stars shown in that snapshot. The current desktop version is v1.5.8, published November 2, so today's signal is renewed attention rather than a new release.
- Repository: zhouxiaoka/autoclip
- Product site: AutoClip
- v1.5.8 release
- CLI and MCP guide
The pipeline, not just the clipper
AutoClip's workflow has five stages:
- Import: use a local file or supported YouTube/Bilibili link.
- Transcribe: use an existing SRT, local transcription or a configured cloud service.
- Find moments: analyse the full transcript for coherent questions, answers and viewpoints rather than cutting at fixed intervals.
- Package: create platform-specific vertical layouts, subtitles, covers and publishing copy.
- Render: use the local FFmpeg pipeline to create the final files.
The project documents platform IDs including Douyin, Xiaohongshu, Bilibili, TikTok, Instagram Reels, YouTube Shorts and YouTube long-form. A podcast crop is not automatically right for an interview, and a short clip that loses the question can be polished but contextually useless.
Output from AutoClip's first-run case study, not a screenshot of the desktop interface. The project says the source tutorial and synthetic narration were created for this demonstration; see the case-study notes.
One pipeline, three entry points
The desktop app, CLI and MCP server share the same production workflow and data directory. A CLI-generated project can appear in the desktop app's project list, rather than becoming a separate automation artifact.
Use the desktop app when a person needs to review clips, the CLI for repeatable batch jobs, and MCP when an agent needs to orchestrate the same workflow. The MCP server uses stdio, and the project also ships an agent skill that explains when to call its tools and how to present results.
Install the CLI
For a source installation, the current developer guide documents Python 3.11+ and FFmpeg:
git clone https://github.com/zhouxiaoka/autoclip.git
cd autoclip
python3 -m venv .venv
source .venv/bin/activate
python -m pip install -r requirements.txt
python -m pip install --no-deps .
autoclip --version
autoclip doctor
On macOS, install FFmpeg with Homebrew if it is not already available:
brew install ffmpeg
The desktop installers bundle portable Python and FFmpeg. The current desktop release supports Apple Silicon Macs on macOS 13+ and Windows 10/11 x64. There is no Intel Mac or Linux desktop installer; use CLI or Docker on those systems.
Produce clips from a script
If subtitles already exist, the first test can avoid a transcription step:
autoclip produce talk.mp4 \
--srt talk.srt \
--platform douyin \
--platform youtube_shorts \
--json
The command returns a project ID and output paths. Export the publishing kits after the job completes:
autoclip outputs PROJECT_ID --export-kits
Replace the example file names and project ID with values from your run. For agent integration, the documented setup is:
uvx autoclip-mcp
npx skills add zhouxiaoka/autoclip
The first command starts the stdio MCP server; the second installs the agent skill. Keep model credentials in environment variables or the app's saved settings rather than pasting them into shell history or a committed config file.
Local processing does not mean every model call is local
Editing and rendering are local, but optional cloud features have different data paths:
- Cloud analysis can send relevant transcript text and captions.
- Optional visual understanding or reference-frame generation can send sampled frames.
- Cloud transcription sends audio to the selected provider.
- Local analysis and local transcription avoid those corresponding cloud calls, but require compatible hardware and model setup.
The project documents Ollama and LM Studio presets. A local model can avoid a cloud model bill, but a smaller model may select weaker clips than a larger hosted model. AutoClip itself is MIT-licensed and free to run; cloud model, transcription and image-generation providers may charge separately.
Review the output before publishing
The project's first-run case study reports processing a 2-minute-7-second original tutorial with an existing SRT into three vertical Shorts in about 62.4 seconds. It explicitly warns that this is not a benchmark for a two-hour interview and excludes setup and review costs.
The same case study documents real output defects: one description ended mid-sentence and another contained an incomplete phrase and shortened tags. That is a useful reality check. Watch every clip, verify that the question and answer remain together, inspect speaker framing and subtitles, and correct the title and description before publishing. A completed render is not an approved post.
There is also an installation caveat: the v1.5.8 release page says the macOS package is not notarised and the Windows installer is unsigned. Download only from the official release page and treat any operating-system warning as a trust decision, not a routine button to click.
AutoClip versus other workflows
| Workflow | Best fit |
|---|---|
| AutoClip | Local rendering, open-source pipeline, CLI/MCP automation and platform-specific output |
| CapCut or a desktop editor | Manual creative control and timeline-level finishing |
| Descript or hosted clippers | Managed transcription and editing workflows |
| FFmpeg + Whisper | Maximum control and a scriptable pipeline, with more engineering effort |
AutoClip is most interesting when you want repeatable production without handing the entire editing workflow to a hosted SaaS. It is less suitable when you need a collaborative editing suite, guaranteed perfect highlight selection or a fully hands-off publishing pipeline.
Official workflow demo
No exact-project YouTube walkthrough was verified in this run. The project publishes a 55-second workflow demo and a 30-second product promo on its own site.
Sources and further reading
- GitHub repository and MIT licence
- AutoClip v1.5.8 release
- CLI, MCP and local-model guide
- Official first-run case study
- Project site and model overview
- Privacy policy
Start with a short recording that already has subtitles and review every generated output. Would you run this as a local CLI job, an MCP tool for an agent, or keep the final editing step in a human-operated editor?
Written by Sh Raj Β· LinkedIn Β· StackShade on YouTube

Top comments (0)