← Back to directory
C

ClipForge MCP Server

Community
Drive ClipForge's keyless short-video pipeline from any MCP client — one sentence or a product link becomes a finished vertical video.
GitHub source repository ↗
★ 756 Stars Category · Other Very popular
58FMRS · C

ClipForge MCP Server is a community-maintained open-source MCP layer that exposes ClipForge's keyless short-video pipeline (script → free stock footage → free TTS → captions → BGM → FFmpeg compose) to any MCP client. It produces videos without API keys and can generate scripts once an LLM is configured; it is self-hosted, watermark-free and AGPL-3.0. It suits users who want local automated production of vertical selling videos driven by AI assistants, and is not for those needing official support or unwilling to run a local instance and FFmpeg.

Reliability
9/20
Security and permissions
10/20
Maintenance
14/20
Documentation
13/20
Setup experience
12/20
Read the FMRS scoring method →

ClipForge (formerly "daihuo-jianshou") is an open-source AI e-commerce and UGC short-video generator: upload a product image and it extracts selling points, writes a script, keeps the product image undistorted, and adds footage, voiceover and captions to produce a vertical selling video for Douyin Shop, Kuaishou, Xiaohongshu, WeChat Channels and TikTok Shop; it also supports one-sentence topic videos. Its MCP server (npm package clipforge-mcp, version 0.1.4) exposes that pipeline over stdio (script → free stock footage → free TTS → captions → BGM → FFmpeg compose) to any MCP client. Footage and voiceover work keyless without an LLM, while script generation requires an OpenAI-compatible LLM endpoint. The project is AGPL-3.0 licensed, self-hosted, and keeps data on the local machine. This MCP server is maintained by the project author as a community open-source effort, not an official offering from any platform or model vendor.

Tools

create_video
Create a short video directly from one sentence or a product link; supports wait:false to submit without blocking and avoid client timeouts.
compose
Run FFmpeg composition on ready script and assets to render the final video; supports wait:false non-blocking submission.
clipforge_update_shots
Refine a single shot with one sentence, using gate/QC feedback for pinpoint edits instead of regenerating the whole script.
clipforge_product_script
Paste a product link to generate a selling script directly.

Setup

  1. Start a ClipForge instance: run docker run -d -p 3000:3000 -v clipforge-data:/data ghcr.io/xixihhhh/clipforge, or run pnpm dev / pnpm start inside the repository.
  2. Add clipforge-mcp to your MCP client config: command npx, args -y [email protected], transport stdio.
  3. Set environment variables as needed: CLIPFORGE_BASE_URL pointing to the running instance (default http://localhost:3000); for AI script writing also set CLIPFORGE_LLM_BASE_URL / CLIPFORGE_LLM_API_KEY / CLIPFORGE_LLM_MODEL (OpenAI-compatible endpoint).
  4. Reload the client, confirm the MCP server is connected, then call the tools to generate videos.
claude_desktop_config.json
{
  "mcpServers": {
    "clipforge": {
      "command": "npx",
      "args": ["-y", "[email protected]"],
      "env": {
        "CLIPFORGE_BASE_URL": "http://localhost:3000",
        "CLIPFORGE_LLM_BASE_URL": "",
        "CLIPFORGE_LLM_API_KEY": "",
        "CLIPFORGE_LLM_MODEL": ""
      }
    }
  }
}

Fit and risk

Best for

  • Teams that need self-hosted, watermark-free short-video production with data staying local
  • Developers who want AI assistants (e.g. Claude Desktop, Cursor) to drive the video pipeline
  • Operators producing e-commerce, UGC ad or faceless vertical content
  • Individuals who want a zero-key path using free footage and free TTS first

Not for

  • Enterprise production environments needing official commercial support or SLAs
  • Users expecting high-quality AI-generated visuals and native voiceover with no API keys at all
  • Users unwilling to run FFmpeg, Node or Docker locally
  • Users who want a fully managed cloud service with no local instance

Required permissions

  • Launch a local npx process (clipforge-mcp) over stdio
  • Read environment variables CLIPFORGE_BASE_URL, CLIPFORGE_LLM_BASE_URL, CLIPFORGE_LLM_API_KEY (secret), CLIPFORGE_LLM_MODEL
  • Access the HTTP API of the locally running ClipForge service (default http://localhost:3000)
  • Make network requests to the configured LLM endpoint when set, for script generation
  • Use FFmpeg on the ClipForge instance for local video composition and read/write local data files

Risks and side effects

  • CLIPFORGE_LLM_API_KEY is a secret credential; leaking it lets others consume your model quota
  • When external LLM endpoints or paid AI models are configured, script and asset calls may incur usage-based costs; paid generation is designed not to trigger silently
  • AI-generated content faces platform compliance rules (AIGC labeling, advertising-law banned words); the project enables these defaults but does not replace your compliance responsibility
  • Free footage comes from third-party licensed libraries (e.g. Openverse, Wikimedia, Jamendo); verify each license and keep attribution before commercial use
  • If the local instance is exposed to a network, others may call it and consume your model quota

Troubleshooting

  1. If tool calls fail, first confirm the ClipForge instance is running and CLIPFORGE_BASE_URL is reachable (default http://localhost:3000)
  2. If script generation errors, check CLIPFORGE_LLM_BASE_URL / API key / model id; note that some platforms use a different chat endpoint path than their asset gateway
  3. Client timeouts can be avoided by using wait:false on compose / create_video for non-blocking submission
  4. If composition is empty or fails, check that FFmpeg is installed locally and that assets downloaded and passed decode validation
  5. For poor output, use clipforge_update_shots for pinpoint shot-level fixes rather than regenerating the whole script

Use cases

Generate a vertical short video from a one-sentence topic inside an MCP client
Turn a product link or product image into a host-style or drama-style selling video
Batch or scheduled (via cron) generation of draft videos for publishing
Import a written script, auto-split into shots, attach free footage and free TTS, and compose
Dubbing and re-composing a script into another language for overseas markets

Supported clients

Claude DesktopFull support
CursorFull support