AutoClip skill
A skill and MCP server for cutting long videos: the agent takes a link or file and builds Shorts, Reels and TikTok clips with a cover, post text and ZIP
- 9,663 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- Runs FFmpeg and downloads video from links, and writes clips, covers and ZIPs into the data directory
- Holds a model key and sends subtitles, text, frames or audio to a cloud provider
- The MCP tools include `publish_clip`, which posts clips through Upload-Post to TikTok, Instagram, YouTube and other accounts
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/autoclipDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Runs FFmpeg and downloads video from links, and writes clips, covers and ZIPs into the data directory
- Holds a model key and sends subtitles, text, frames or audio to a cloud provider
- The MCP tools include `publish_clip`, which posts clips through Upload-Post to TikTok, Instagram, YouTube and other accounts
This is third-party code. Review the repository files before installing.
What it does
AutoClip is long-video clipping driven by an agent. The `autoclip` skill teaches Claude Code, Cursor or Codex to turn a YouTube or Bilibili link, or a lecture, interview, podcast or course recording from disk, into ready videos for the platforms you pick. The same jobs show up in the AutoClip desktop app: it shares a data directory with the `autoclip` command and the MCP server. The agent calls `start_quick_output` and gets a `project_id` straight away; without MCP it runs `autoclip produce` in the shell. It then polls `get_quick_output_status` at the `poll_after_sec` interval until the job reaches a final status: `completed`, `partial`, `failed`, `interrupted` or `cancelled`. Inside, the service takes subtitles (the video's own or ones recognised by local Whisper), hands the text to the chosen model to find strong moments, cuts the video with FFmpeg and builds a version for each platform. Each platform gets a video, a cover, post text and a ZIP kit. Platforms are set with the identifiers `douyin`, `xiaohongshu`, `bilibili`, `tiktok`, `instagram_reels`, `youtube_shorts` and `youtube_long`. The editing style is one of `editorial`, `street` and `classic`, and the vertical layout one of `auto`, `interview` and `podcast`. The agent submits links and files one at a time and ends with a table: source, `project_id`, status and kit path.
Who it is for. For podcast and course creators, social media managers and marketers who need to cut long recordings into short-video clips and hand that work to an agent.
Good fit when
- You have a two-hour interview or lecture and need vertical clips for Shorts, Reels or TikTok with a cover and post text
- A podcast comes out every week and you want an agent to do the cutting: a link in the request, a ZIP kit in return
- One episode has to go to several platforms at once, for example `douyin` and `youtube_shorts`, with Chinese and English versions
- You need the same result from a script on the command line: `autoclip produce` returns JSON with paths to the video, cover and text
- You want to look over the clips in the desktop app first, then adjust them and download the kit
Not a fit when
- You only want a list of scored fragments and compilations without editing and covers: the repository keeps the older mode in `references/legacy.md` for that, and the skill points there
- You need to run a dozen links in parallel: the skill's own rules take them one at a time, and there is no parameter for a list of sources
- The video has hardly any speech, for example a game stream: the verified cases are mostly interviews, podcasts, courses and talks, and games need frame analysis with an image-capable model
- You have an Intel Mac and want the desktop app itself: builds exist for Apple Silicon and Windows x64, and everything else runs through Docker or the command line
Example request
Turn this podcast episode https://www.youtube.com/watch?v=VIDEO into clips for youtube_shorts and tiktok in the editorial style and show me the paths to the videos, covers and ZIPLimitations
The skill itself and the CLI and MCP reference are written in Chinese; the README exists in Russian and English translations. The agent reads the version of the installed package with `autoclip --version` or the `get_version` tool: the documents do not pin a number.
The command line and MCP need Python 3.10 or newer (the author recommends 3.11) and FFmpeg with FFprobe on PATH. The model comes from the desktop app's settings or the `AUTOCLIP_API_KEY` variable; local Ollama and LM Studio need no key. A video without subtitles needs local Whisper or cloud transcription.
The `youtube_long` platform builds only whole fragments of 180 seconds or longer. The `template` value is accepted while the `mcp_v2_tools` flag is on.
One person maintains the project and there is no live support.
How to disable. Remove the skill the way your agent does it, for example delete the `autoclip` folder from `~/.claude/skills/` or `~/.cursor/skills/`, and the MCP server with `claude mcp remove autoclip`. Clips and projects stay in the app's data directory and are deleted by hand.
FAQ
What is AutoClip and how do I connect it to Claude Code?
AutoClip is an open-source video clipping app with an `autoclip` command, an MCP server and a skill for agents. For Claude Code, add the MCP with `claude mcp add-json autoclip '{"command":"uvx","args":["autoclip-mcp"],"env":{"AUTOCLIP_API_KEY":"${AUTOCLIP_API_KEY}"}}'` and install the skill with `npx skills add zhouxiaoka/autoclip`. The single quotes leave the variable unexpanded in the config, so the key does not land in `~/.claude.json` as plain text.
Is AutoClip free and does it need an API key?
The app is free and MIT-licensed. Cloud analysis, cloud transcription and AI images run on your keys at the provider's rates. Auto cover does not call paid generation by default. Local Ollama and LM Studio need no key, only suitable hardware.
Where does my video go during clipping?
Cutting and rendering happen on your computer. The cloud model receives the subtitles and text it needs, image analysis the necessary frames, and cloud transcription the audio track. If both analysis and speech recognition are local, nothing leaves the machine.
Why does CLI and MCP produce a classic clip while the app gives an editorial one?
It is by design. The app offers `editorial` by default, but `produce` and `start_quick_output` without a style use `classic`: identical calls give identical results before and after an update. To get the app's look, pass `--template editorial` or `template: "editorial"`.
A job finished with the partial status, what now?
The `partial` status means some versions failed; the finished ones are kept. By the skill's rules the agent hands over what is done, names the failed platforms and does not resubmit the whole order. The `interrupted` status means the worker process ended early: finished videos can be fetched and the rest is the user's call.
Will a short video work with youtube_long?
No. `youtube_long` builds only whole fragments of at least 180 seconds, so for a short source choose `youtube_shorts` or `bilibili`.
Discussion
No questions or reviews yet.
Related
SkillsA Claude Code skill that downloads a YouTube video, splits it into semantic chapters, cuts chosen clips, translates subtitles and burns bilingual ones in
A skill that pulls highlights from a long video: download, transcript, segmentation, analysis and clips with titles
ffmpeg-skill — a skill and MCP server: an agent cuts and joins video, fixes audio, adds captions and exports for TikTok and Reels with local FFmpeg
An MCP server and REST API to assemble short vertical videos with captions, voiceover and music for TikTok, Reels and Shorts
ChatGPT short-video editing skills
ChatGPT 剪短影音 Skills
Two agent skills: set up the editing environment and build a vertical short video in eight verified steps
A video editing skill for Claude Code and Codex: the agent cuts pauses and filler words, grades color, burns subtitles and adds animated overlays
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Repository stars | Main difference |
|---|---|---|---|
| AutoClip (this entry) | no | 9,663 | |
| OpenClip | no | 575 | finds and cuts the best moments of long videos and streams with titles and covers, works as a skill |
| YouTube Clipper Skill | no | 2,254 | splits a YouTube video into topical chapters, translates subtitles and burns bilingual ones into the clip |
| SeeCut | no | 378 | a Claude Code skill that turns a talking-head video into a short |
README in short
The front page promises one click from link to video, cover and text for Douyin, Xiaohongshu, TikTok, Reels, YouTube Shorts, Bilibili and YouTube, with sixteen finished example clips attached. Next come a table of time and cost on three real recordings, a section on data flow, a quick start for the desktop app, Docker and CLI with MCP, and a list of models. The repository holds guides for installation, Docker and troubleshooting, a changelog, a roadmap, a privacy note, the project description in eight languages and backend tests, including a check of the skill contract.
README badge
Original SKILL.md
--- name: autoclip description: >- 用 AutoClip 把一条 YouTube / B 站链接，或本地长视频（讲座、访谈、播客、课程），做成各平台成片、封面、发布文案和 ZIP 发布包。 当用户要「出片」「做封面」「写发布文案」「打包发布」「从这个链接做短视频」时使用。 只要原始高光切片、合集，或把旧切片发到 Upload-Post 时不要用这份说明，改读 references/legacy.md。 --- # AutoClip：从链接或长视频到发布包 AutoClip 在本机把一条素材做成可发布的成片。产物和桌面应用共用数据目录，桌面首页能看到同一个项目。 先调用 `get_version`（或 `autoclip --version`）读取当前安装的版本。版本号来自安装包，不要和文档里的数字核对。 ## 什么时候用 用户给出下面任一输入，并且要的是成片而不是素材列表： - HTTPS 的 YouTube 或 B 站链接 - 本地长视频的绝对路径（讲座、访谈、播客录像、课程） 交付物是四样，缺一要说明哪一样没有： 1. 按平台剪好的成片 2. 封面 3. 发布文案 4. ZIP 发布包 ## 什么时候不用 - 只要带评分的原始切片和合集：读 [references/legacy.md](references/legacy.md)。 - 把已经切好的旧片段发到 TikTok / YouTube 等（Upload-Post）：同样读那份参考，不要走下面的出片流程。 - 用户只是在问 AutoClip 怎么安装、或还没有视频：先说明装法，不要空跑任务。 ## 先选调用方式 1. **有 AutoClip MCP 工具** → 用工具，不要自己拼命令。 2. **没有 MCP，但能跑 shell** → `autoclip produce <source> --json`。 3. **两者都没有** → 一行安装：`uvx autoclip-mcp`，skill 用 `npx skills add zhouxiaoka/autoclip`。备选是在仓库根目录执行 `python -m pip install -r requirements.txt`，再执行 `python -m pip install --no-deps .`。详见 `docs/CLI_AND_MCP.md`。 模型密钥放在桌面应用的设置里，或环境变量 `AUTOCLIP_API_KEY`。不要把 key 写进工具参数或对话。 ## 示例 1：链接出发布包 1. `get_version`（或 `autoclip --version`），记下返回的版本。不要对照文档里写死的版本号。