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
- 1,127 ★ on the repository
- Last commit
- AGPL-3.0
Extended access
Why this level
- Composes video through a local service and FFmpeg, which run code on your machine
- The ingest_product tool fetches whatever product link it is given
- Sends the model key to an external OpenAI-compatible service on every script generation
Install
- Works in
Manual install
claude mcp add clipforge -- npx -y --package https://github.com/xixihhhh/clipforge/releases/download/v0.9.7/clipforge-mcp-0.1.6.tgz clipforge-mcpA route with no repository clone: installs the version from a specific GitHub release. Set CLIPFORGE_LLM_BASE_URL, CLIPFORGE_LLM_API_KEY, CLIPFORGE_LLM_MODEL in the MCP server's env, and run ClipForge itself (pnpm dev or pnpm start) separately.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Composes video through a local service and FFmpeg, which run code on your machine
- The ingest_product tool fetches whatever product link it is given
- Sends the model key to an external OpenAI-compatible service on every script generation
This is third-party code. Review the repository files before installing.
What it does
ClipForge MCP is a thin wrapper over the ClipForge application's HTTP API that Claude Desktop, Claude Code and Cursor call as a regular MCP server. The clipforge_create_video tool takes a topic in any language, writes the narration, pulls free stock video and photos from Wikimedia Commons, voices the script with Edge TTS and composes a vertical video with FFmpeg, with no paid key at all. clipforge_ingest_product takes a product link, pulls its title, price and photo from the page's markup and creates a project from that data right away. Before every compose, the narration passes through five built-in reviewer roles that break it down shot by shot, rewrite weak lines and check that a price or spec number survives the rewrite unchanged. A finished video is not treated as done on its own: the agent has to call clipforge_master to check splice continuity and loudness, then clipforge_gate, which rolls script, quality and footage-licensing checks into one verdict, and clipforge_contact_sheet to look at the frames at splice points before handing over the file link. Script generation needs one OpenAI-compatible model key; the rest of the server's tools, including stock search, the list of Edge TTS voices and export to OTIO or EDL for editing software, work key-free. The MCP server can be started key-free through an npx package from a GitHub release, but it still needs a locally running ClipForge instance with FFmpeg.
Who it is for. For marketplace sellers and short-video creators who want a finished clip from a topic or a product straight out of a chat with an agent, without a paid video-generation subscription.
Good fit when
- You need a vertical clip on a topic for TikTok, Reels or YouTube Shorts without a paid generation service
- You have a product link and want it turned into an ad video with the price and photo, no manual editing
- You want a finished script, voiceover and captions in one call from Claude Code or Cursor
- You want to export the edit plan to OTIO or EDL for a manual pass in a professional editor
Not a fit when
- There is no locally running ClipForge instance with FFmpeg installed: the server only forwards calls to it
- You need an on-camera talking presenter or lip sync: the free path gives a voiceover and stock footage, not a digital avatar
- You have no OpenAI-compatible model key: without one, clipforge_create_video cannot write the script
Example request
Make a vertical video about how to brew pour-over coffee, with free voiceover and captionsMCP
- Transport
- stdio
- Authentication
- not required
Environment variables
CLIPFORGE_BASE_URL- Address of the running ClipForge instance, defaults to http://localhost:3000
CLIPFORGE_LLM_API_KEYrequired, secret- OpenAI-compatible model key, needed only for script generation
Limitations
License is AGPL-3.0: distributing a modified version as a service means releasing that source too, so check the terms with the author before commercial use.
- The MCP server does not compose video itself: it needs a separately running ClipForge instance (`pnpm dev` or `pnpm start`) with FFmpeg installed on the same machine.
- The server package is not yet published to npm or the MCP registry; running it without cloning the repository goes through a specific GitHub release archive.
- Free video and photos come from Wikimedia Commons; better-quality stock needs separate Pexels or Pixabay keys.
- Script generation needs an OpenAI-compatible model key and spends its quota on every clipforge_create_video or clipforge_generate_script call.
- Edits to an existing video go only through a preview plan and confirmation step: applying a change without showing the diff to the agent first breaks the server's own rules.
How to disable. Remove the clipforge entry from the MCP client configuration (claude mcp remove clipforge in Claude Code, or the mcpServers block in Claude Desktop and Cursor). If the server ran through the release's npx package, no files remain besides the npx cache.
FAQ
Do I need a paid key to make a video with ClipForge MCP?
Not for footage or voice: video and photos come from Wikimedia Commons and Edge TTS voices the narration, both key-free. A paid key is only needed for the OpenAI-compatible model that writes the script, spent on every clipforge_create_video call.
Can I connect ClipForge MCP without cloning the repository?
Yes, through a specific GitHub release's archive, via npx with --package and a direct .tgz link. The server still sends requests to a running ClipForge instance, so one has to be started separately.
How does the server turn a product link into a video?
clipforge_ingest_product reads the product page's markup (JSON-LD, then Open Graph, then a plain title) and pulls the name, price and photo. That data seeds a project, and clipforge_product_script then writes the script.
What do the five built-in reviewer roles check before composing?
They break the narration down shot by shot and flag functional phrases like 'shows the product's effect', asking for a visible action instead. A rewritten line goes through a separate check: if it drops a price or spec number from the original phrase, the whole line is discarded.
Why shouldn't the agent hand over the finished video right away?
The server's own rules require looking at the frames at splice points with clipforge_contact_sheet first: automated checks cannot see overlapping captions or an ugly frame, but the image shows it right away.
Discussion
No questions or reviews yet.
Related
MCP serversMCP server, CLI and desktop app that cut a long video or stream replay into vertical clips with captions
Claude Video Vision
claude-video-vision
An MCP server and Claude Code plugin that let the agent watch video: frames via ffmpeg and audio transcription via Gemini, local Whisper or OpenAI
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
VidMoat MCP
Vidmoat MCP
Remote MCP server of a cloud video editor: an agent imports footage, edits the timeline, checks frames and renders MP4
Skills, a CLI and an MCP server that let Claude Code or Codex make video from a brief: editing, graphics, voiceover, subtitles
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| ClipForge MCP (this entry) | yes | not required | 1,127 | |
| short-video-maker | no | not required | 1,443 | renders a clip from scenes you define through its own Docker image, without generating a script from a topic |
| MCP for publishing short video to 8 platforms | no | API key | 1 | publishes a finished vertical video straight to eight platforms, but does not compose the video itself |
README in short
The description, in Chinese and English, opens with a comparison table against other tools on agent access: MCP, a command line, a skill and batch composition in one place. Sections on frame-composition control, a version-numbered changelog and the license follow. A closing 'Tech stack' section lists a Next.js and React front end, a route-handler API layer and test coverage run through GitHub Actions.