freemcplab / 目錄 / 演練場 / Claude Video Vision

● 社群 jordanrendric ⚡ 即開即用

Claude Video Vision

Name: Claude Video Vision (Claude Skill)
Author: jordanrendric

作者 jordanrendric · jordanrendric/claude-video-vision

Make Claude actually understand a video — frames + transcript + scene segmentation, fed in a token-efficient layout.

The skill takes a video file or URL, extracts keyframes (scene-aware), aligns transcript segments, and presents the bundle to Claude for QA, summary, or content extraction. Smart-sampling avoids blowing the context window.

▶ 觀看演示 📦 安裝 💡 使用場景

為什麼要用

核心特性

Scene-aware keyframe extraction
Whisper-quality transcript alignment
Per-scene token budget
Works with local files or YouTube URLs
Output: structured JSON for downstream skills

即時演示

實際使用效果

就緒

安裝

選擇你的客戶端

~/Library/Application Support/Claude/claude_desktop_config.json · Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "claude-video-vision": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/jordanrendric/claude-video-vision"
      ],
      "_inferred": true
    }
  }
}

開啟 Claude Desktop → Settings → Developer → Edit Config。儲存後重啟應用。

~/.cursor/mcp.json · .cursor/mcp.json

{
  "mcpServers": {
    "claude-video-vision": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/jordanrendric/claude-video-vision"
      ],
      "_inferred": true
    }
  }
}

Cursor 使用與 Claude Desktop 相同的 mcpServers 格式。專案級設定優先於全域。

VS Code → Cline → MCP Servers → Edit

{
  "mcpServers": {
    "claude-video-vision": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/jordanrendric/claude-video-vision"
      ],
      "_inferred": true
    }
  }
}

點擊 Cline 側欄中的 MCP Servers 圖示，然後選 "Edit Configuration"。

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "claude-video-vision": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/jordanrendric/claude-video-vision"
      ],
      "_inferred": true
    }
  }
}

格式與 Claude Desktop 相同。重啟 Windsurf 生效。

~/.continue/config.json

{
  "mcpServers": [
    {
      "name": "claude-video-vision",
      "command": "git",
      "args": [
        "clone",
        "https://github.com/jordanrendric/claude-video-vision"
      ]
    }
  ]
}

Continue 使用伺服器物件陣列，而非映射。

~/.config/zed/settings.json

{
  "context_servers": {
    "claude-video-vision": {
      "command": {
        "path": "git",
        "args": [
          "clone",
          "https://github.com/jordanrendric/claude-video-vision"
        ]
      }
    }
  }
}

加入 context_servers。Zed 儲存後熱重載。

claude mcp add claude-video-vision -- git clone https://github.com/jordanrendric/claude-video-vision

一行命令搞定。用 claude mcp list 驗證，claude mcp remove 移除。

使用場景

實戰用法： Claude Video Vision

Answer questions about a 30-min talk

👤 Claude Code users ⏱ ~15 min intermediate

何時使用： Conference talk just dropped; you want notes without watching it.

步驟

Ingest

Point skill at the URL.✓ 已複製

→ Frames + transcript ready
Ask

Ask focused questions; skill scopes to relevant scenes.✓ 已複製

→ Precise answers with timestamps

結果： Targeted understanding without re-watching.

組合

與其他 MCP 搭配，撬動十倍槓桿

claude-video-vision + filesystem

Save the structured note output for later reference

Combine claude-video-vision with filesystem: Save the structured note output for later reference✓ 已複製

工具

此 MCP 暴露的能力

工具	輸入參數	何時呼叫	成本
ingest	(see docs)	Pull video + transcript	1 call
query	(see docs)	Scope-aware question answering	1 call

成本與限制

運行它的成本

API 配額: Bound by transcription cost
每次呼叫 Token 數: Mid (sampled, not full video)
費用: Free skill; transcription/model costs
提示: Use shorter scenes when QA is targeted; use long-window mode for summary

安全

權限、密鑰、影響範圍

最小權限： fs-read fs-write

憑證儲存： API keys for transcription provider in env

資料出站： Transcription provider

切勿授予： process private video without consent

故障排查

常見錯誤與修復

Transcript misaligned

Lower scene-segmentation threshold; transcript may be off due to music

替代方案