Smart Clip MCP

AI-powered smart video clipping MCP server. Input a long video + editing intent, output highlight short clips.

Not another FFmpeg wrapper — it's the "editing brain". Uses subtitle semantic analysis + LLM-driven decision making to identify highlight moments, with KyaniteLabs/mcp-video as the execution layer.

Features

• 🧠 LLM-driven highlight detection — analyzes subtitles to identify the most engaging moments

• 🎬 5 MCP tools — smart_clip, repurpose, highlight_reel, analyze_content, get_edit_plan

• 🎯 Platform-adaptive — auto-resize and format for TikTok, YouTube Shorts, Instagram Reels

• 📝 Auto subtitles — Whisper transcription + burn-in with platform-specific styling

• 🔊 Audio analysis — energy peaks and silence detection for precise cut points

• 👀 Human-in-the-loop — preview edit plans before execution

• 💰 Low cost — ¥0.8-1.16 per hour of video (50x cheaper than cloud alternatives)

Related MCP server: podcli

Quick Start

Prerequisites

• Python 3.11+

• FFmpeg installed and on PATH

• Whisper model (auto-downloaded on first use)

Install

pip install smart-clip-mcp

Configure MCP Client

Claude Code:

claude mcp add smart-clip -- uvx --from smart-clip-mcp smart-clip-mcp

Claude Desktop / Cursor:

{
  "mcpServers": {
    "smart-clip": {
      "command": "uvx",
      "args": ["--from", "smart-clip-mcp", "smart-clip-mcp"],
      "env": {
        "OPENAI_API_KEY": "sk-..."
      }
    }
  }
}

Usage

Ask your AI agent:

"Extract 5 highlight clips from this 1-hour podcast video"

"Turn this interview into 3 TikTok-ready shorts"

"Analyze this video and tell me the most engaging moments"

MCP Tools

Architecture

Video → [Analyzer] → [Planner] → [Executor] → Clips
          │              │            │
          │ Whisper       │ LLM        │ mcp-video
          │ librosa       │ Prompts    │ FFmpeg
          │ PySceneDetect │ Strategy   │

• Analyzer — Content understanding: Whisper transcription, audio energy analysis, scene detection

• Planner — Decision making: LLM highlight detection, template matching, strategy engine

• Executor — Clip generation: trim, merge, subtitles, platform adaptation via mcp-video

Configuration

Create ~/.smart-clip/config.yaml:

analyzer:
  whisper:
    mode: local          # local | api
    model: large-v3
    language: zh
  audio:
    energy_percentile: 90
    silence_threshold: 0.3

planner:
  llm:
    model: gpt-4o-mini
    temperature: 0
  strategy:
    min_score: 6.0
    min_gap: 10

executor:
  output:
    format: mp4
    quality: high

Development

# Clone
git clone git@github.com:Ambrose1/Smart-Clip-MCP.git
cd Smart-Clip-MCP

# Create venv
python -m venv .venv
source .venv/bin/activate

# Install dev dependencies
pip install -e ".[dev]"

# Run tests
pytest

# Run MCP server locally
smart-clip-mcp

License

Apache 2.0 — see LICENSE.