CutPilot

CutPilot

Local-first AI video editing engine controlled by MCP-compatible AI hosts, enabling AI to inspect footage, propose edits, modify multitrack projects, and render locally while keeping user control.

Category
访问服务器

README

CutPilot 10

English | 简体中文

CutPilot is a local-first AI video editing engine with its own configurable AI mode, plus integrations for Codex, Claude, and any MCP-compatible host. The AI can inspect footage, propose reviewable edits, modify a real multitrack project, open an embedded timeline, and render locally while the user keeps control of source media and every edit decision.

Version 10 adds a persistent project-wide semantic asset index, a review-first cross-category Director Agent, and a durable background task center with cancellation, retry, and restart recovery. It matches narration beats to exact local source ranges, avoids repeated shots, and writes only approved plans to an editable timeline.

Highlights

  • 219 MCP tools, including four high-level workflow entries, and nine project starters.
  • Independent AI planning through OpenAI, Anthropic, or a compatible endpoint; every timeline mutation still requires approval.
  • Atomic project writes, cross-process locking, automatic backup recovery, schema migration, repair, Undo/Redo, and snapshots.
  • Multitrack video/audio timelines, linked A/V, trimming, splitting, ripple operations, snapping, markers, keyframes, effects, transitions, captions, Undo/Redo, and background exports.
  • Local transcription, transcript editing, silence detection, scene detection, beat analysis, duplicate-shot review, proxy media, and smart reframing.
  • Automatic asset understanding using FFmpeg evidence and Apple Vision semantic labels on macOS. Person identity is never inferred.
  • Review-first natural-language edits for speed, volume, captions, mute, deletion, fades, dissolves, scale, exposure, saturation, and track locking. Open-ended creative intent can be decomposed by CutPilot's configured AI or a connected host.
  • Vlog, talking-head, podcast, wedding, product-promo, explainer, and motion-graphics director workflows.
  • SVG and JSX/React motion graphics, WebGL shaders, concurrent shader batches, and real Remotion 4 project rendering.
  • FCPXML, Premiere XML, EDL, SRT/VTT/ASS, and a safe CapCut/Jianying continuation bundle.
  • Local procedural media and macOS voice generation plus configurable OpenAI, Seedance, Kling, Mureka, SFX, and generic HTTP bridges.

What CutPilot is

CutPilot is an AI editing backend with an embedded manual review interface. It is not intended to replace a traditional desktop editor for every hand operation. Its built-in AI, Codex, or Claude can direct the edit, and the user can open CutPilot's local interface to adjust the timeline, captions, effects, audio, and motion graphics.

Original media stays local unless the user explicitly selects a remote generation or transcription provider.

Requirements

  • macOS
  • Node.js 18 or newer
  • FFmpeg and FFprobe
  • Google Chrome
  • ImageMagick for the local image generator
  • Xcode command-line tools for Apple Vision classification

For Homebrew users:

brew install ffmpeg imagemagick
xcode-select --install

One-command product install from a checked-out repository:

./install-macos.sh --install-deps

Launch without Codex or Claude and optionally import a media folder:

~/.local/bin/cutpilot --project ~/Movies/my-video.cutpilot.json --media ~/Movies/clips

Open the AI tab to configure OpenAI, Anthropic, or a compatible endpoint. The API key is stored in ~/.cutpilot/settings.json with user-only permissions and is never written to a project.

Run from source

npm install
npm run validate
node scripts/server.mjs

Example MCP configuration:

{
  "mcpServers": {
    "cutpilot": {
      "command": "node",
      "args": ["/absolute/path/to/cutpilot/scripts/server.mjs"]
    }
  }
}

Ready-made examples for Codex, Claude Code, and Claude Desktop are included in the repository.

Safety model

  • Natural-language mutations are planned first and require explicit approval.
  • Destructive actions are marked in the plan.
  • Dependency installation for external Remotion projects requires approval; lifecycle scripts are disabled by default.
  • Apple Vision may classify scenes and count detected humans/faces, but CutPilot does not identify people.
  • Modern Jianying/CapCut encrypted drafts have no public write protocol. CutPilot provides read-only detection and an editable media/subtitle/timecode handoff rather than claiming decryption.
  • Paid generation models require the user's own account, endpoint, authorization, and billing.

Run the live readiness audit through the MCP tool audit_runtime_readiness to see what is actually available on the current machine.

Validation

npm run test:core
npm run test:browser

The tests exercise real FFmpeg media, Apple Vision, natural-language plans, the embedded browser UI, Remotion rendering, generation jobs, CapCut handoff, WebGL batch rendering, and MCP startup.

Current external boundaries

  • Commercial generation providers are only live after credentials and endpoints are configured.
  • Direct writing of modern proprietary encrypted Jianying drafts remains experimental.
  • Open-ended creative instructions require either a configured CutPilot AI provider or an AI host such as Codex or Claude.

License

No open-source license has been granted yet. The source is publicly visible, but reuse and redistribution remain reserved unless a license is added later.

推荐服务器

Baidu Map

Baidu Map

百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。

官方
精选
JavaScript
Playwright MCP Server

Playwright MCP Server

一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。

官方
精选
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。

官方
精选
本地
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。

官方
精选
本地
TypeScript
VeyraX

VeyraX

一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。

官方
精选
本地
graphlit-mcp-server

graphlit-mcp-server

模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。

官方
精选
TypeScript
Kagi MCP Server

Kagi MCP Server

一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。

官方
精选
Python
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选
Neon MCP Server

Neon MCP Server

用于与 Neon 管理 API 和数据库交互的 MCP 服务器

官方
精选
Exa MCP Server

Exa MCP Server

模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。

官方
精选