HotClip MCP
MCP server, CLI and desktop app that cut a long video or stream replay into vertical clips with captions
- 394 ★ on the repository
- Last commit
- AGPL-3.0
Extended access
Why this level
- Reads video files and writes clips, covers and copy to the chosen folder
- With a cloud model the transcript goes to the address you set
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/hotclipDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Reads video files and writes clips, covers and copy to the chosen folder
- With a cloud model the transcript goes to the address you set
This is third-party code. Review the repository files before installing.
What it does
HotClip turns a podcast, stream replay or course into a set of short vertical 9:16 videos. At its core it is a desktop app, and the same pipeline is open to an agent through a command line, a local MCP server and a skill. Speech recognition runs on the device, and the language model that finds highlights connects through any OpenAI-compatible address, including local Ollama. The MCP server exposes three tools: transcribe_video recognizes speech word by word, detect_highlights proposes candidates with a score, a hook and a reason for the pick, and clip_video runs the full cycle from recognition to finished files. You can review the candidate list before cutting, and the next run reuses the recognition cache. You can pass in a viral clip as a reference, and the project measures its pacing, speech rate and shot-cut frequency locally. For each clip the output holds an mp4 file, a cover and publish copy, and clips.json records a quality report. It flags black frames and long silences, as well as mid-word cuts.
Who it is for. For podcasters, streamers and video editors who want to hand clipping to an agent in Claude Code or Claude Desktop.
Good fit when
- You have a four-hour broadcast recording and want a dozen clips from one sentence to an agent
- A podcast exists only as audio, and platforms need a vertical video with captions
- You first need a candidate list with reasons and will cut after reviewing it
- A clip should follow the pacing and hook of a successful reference you pass as a separate file
- You want unattended clipping: the app can watch a recording folder and cut new files
Not a fit when
- You need fine manual frame and effect layout: this is automatic clipping with checks and has no editing desk
- The computer is too weak for local speech recognition
- You want clips published to platforms straight away: the project prepares files and copy, and you post yourself
- You need to embed the code in a closed product, and AGPL-3.0 does not allow that without disclosing the source
Example request
Find the ten strongest moments in broadcast.mp4, show the list with reasons, and after I reply cut vertical clipsMCP
- Transport
- stdio
- Authentication
- not required
Environment variables
HOTCLIP_LLM_BASE_URLrequired- Address of the OpenAI-compatible LLM API, for example local Ollama
HOTCLIP_LLM_MODELrequired- Model name used to find highlights
HOTCLIP_LLM_API_KEYsecret- Cloud model key, not needed for local Ollama
Limitations
- The code is AGPL-3.0-only; installers exist for Windows and macOS, and the Linux build is marked experimental.
- The MCP server needs a clone of the repository and dependencies installed with pnpm. It starts from the project folder through tsx and is not an npm package.
- Finding highlights needs an OpenAI-compatible model address: local Ollama or a cloud key. Without a configured model the skill asks which endpoint to use and never guesses keys.
- The first recognition run downloads a model of about 1 GB, which takes several minutes.
- The docs are in Chinese and English, and publishing packs target Chinese platforms and TikTok. There is no Russian interface.
- Cloud AI keys and addresses are passed through environment variables. What goes to the cloud depends on the model you choose.
How to disable. Remove the hotclip entry from the MCP client configuration and restart the client. If you installed the skill, delete the ~/.claude/skills/hotclip folder. The cloned repository and downloaded models can be deleted by hand.
FAQ
How do I connect HotClip to Claude Code?
Clone the repository, run `pnpm install` and add an MCP server with the command `npx -y tsx src/mcp/server.ts` and cwd set to the project folder. Set HOTCLIP_LLM_BASE_URL and HOTCLIP_LLM_MODEL in the environment, plus HOTCLIP_LLM_API_KEY for cloud endpoints.
What tools does the HotClip MCP server have?
Three: transcribe_video for speech recognition, detect_highlights for a list of strong moments and clip_video for full clipping. The clipping result is described in clips.json.
Can I review candidates before cutting?
Yes, the skill recommends that order: `pnpm cli highlights video.mp4 --json` shows titles, scores and hooks, and clip then reuses the recognition cache.
What does HotClip do with my video, and does it go to the cloud?
Recognition, cutting, captions and export run locally and source recordings are not uploaded. Only the language model address can be external: with local Ollama the work stays fully offline.
Which file formats work?
MP4, MKV, MOV, FLV and TS, plus audio-only files: podcasts get an auto-generated waveform video. Paths with spaces must be quoted.
Discussion
No questions or reviews yet.
Related
MCP serversA 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
ClipForge MCP
ClipForge
ClipForge MCP builds a vertical video from a topic or a product link: stock footage, Edge TTS voiceover and captions, no keys, composed with FFmpeg
An MCP server for video editing built into the free Concat editor: the agent cuts clips, adds titles and captions, sets keyframes and exports video
An MCP server and REST API to assemble short vertical videos with captions, voiceover and music for TikTok, Reels and Shorts
MCP server that gives an agent a YouTube video's transcript from its link, with language choice, timestamps and ads filtered out
A skill that pulls highlights from a long video: download, transcript, segmentation, analysis and clips with titles
README in short
The core of the project is an Electron desktop app with bundled FFmpeg and local recognition through sherpa-onnx. Beyond clipping it watches a recording folder, weighs live chat from stream recordings and tunes picks to the stats of past posts. A separate section covers the CLI, MCP server and skill for agents.