Remotion MCP App
Enables AI-powered video creation by letting the model write React/Remotion code, compiling it in real-time, and rendering the video inline in the chat interface via an interactive widget.
README
<p align="center"> <a href="https://github.com/mcp-use/remotion-mcp-app/releases/download/v1.0.0/ClaudeRemotionDemo.mp4"> <img src="assets/demo-claude.gif" alt="Remotion MCP App Demo" width="100%"/> </a> </p>
Remotion MCP App
An MCP App for AI-powered video creation. It combines an MCP server with an interactive widget — the model writes React/Remotion code, the server compiles it in real-time, and a live video player renders the result directly inside the chat.
Unlike a standard MCP server that only returns text, an MCP App bundles a full UI widget alongside its tools. The Remotion Player widget renders inline in any compatible client (ChatGPT, Claude, or custom apps built with mcp-use), giving the model a visual canvas it can iterate on.
Try it now
Connect to the hosted instance:
https://still-feather-l5mwy.run.mcp-use.com/mcp
Setup on ChatGPT
- Open Settings > Apps and Connectors > Advanced Settings and enable Developer Mode
- Go to Connectors > Create, name it "Remotion", paste the URL above
- In a new chat, click + > More and select the Remotion connector
Requires ChatGPT Pro, Team, Enterprise, or Edu. Full guide
Setup on Claude
- Open Settings > Connectors > Add custom connector
- Paste the URL above and save
- The Remotion tools will be available in new conversations
Requires Claude Pro, Max, Team, or Enterprise. Full guide
Demos
ChatGPT + Remotion MCP App
Claude + Remotion MCP App
Click the previews above to watch the full demos.
How it works
This is an MCP App — an MCP server paired with a UI widget. The two pieces work together:
- MCP Server -- exposes
create_videotool + rule tools for teaching Remotion patterns - Widget -- a Remotion Player that renders inline in the chat, receives compiled bundles from the server
The flow:
- The model calls
create_videowith React/Remotion source files - The server compiles the project with esbuild (sub-second)
- The compiled bundle is sent back as
structuredContent, and the widget renders it as a playable video - For edits, the model calls
create_videoagain with only changed files -- the widget updates in-place with a loading overlay
Model MCP App (Server + Widget)
| |
|-- create_video({files}) ->|
| |-- esbuild compile
|<- structuredContent -----|
| |-- Widget renders video inline
| |
|-- create_video({edits}) ->|
| |-- merge + recompile
| |-- Widget updates in-place
Single tool design
There is one tool: create_video. It handles both creation and editing. The files parameter is always required -- for edits, only send changed files. The server merges them with the previous session state.
Rule tools
The server includes teaching tools derived from the remotion-best-practices skill that the model can call to learn Remotion patterns:
| Tool | Topic |
|---|---|
rule_react_code |
Project structure, imports, entry file contract |
rule_remotion_animations |
useCurrentFrame, frame-driven animation |
rule_remotion_timing |
interpolate, spring, Easing configs |
rule_remotion_sequencing |
Sequence, scene management, duration |
rule_remotion_transitions |
TransitionSeries, fade, slide, wipe |
rule_remotion_text_animations |
Typewriter effect, word highlighting |
rule_remotion_trimming |
Trimming with negative Sequence from |
Widget (the "App" part)
The Remotion Player widget is what makes this an MCP App rather than a plain MCP server. It runs inside the chat interface and features:
- Live video playback with controls
- Animated loading state with shader gradient while the model writes code
- Editing overlay (blur + gradient) when updating an existing video
- Fullscreen mode
- Error display with compilation error details
Local development
Prerequisites
- Node.js 18+
- npm or pnpm
Setup
git clone https://github.com/mcp-use/remotion-mcp-app.git
cd remotion-mcp-app
npm install
npm run dev
The server starts at http://localhost:3000/mcp.
Connect a client
Point any MCP client at http://localhost:3000/mcp. For example with mcp-use:
{
"mcpServers": {
"remotion": {
"url": "http://localhost:3000/mcp"
}
}
}
Build and deploy
npm run build
npm run deploy
Project structure
index.ts -- MCP server, tool definitions, handler
utils.ts -- esbuild compilation, session state, response helpers
types.ts -- Shared types (VideoProjectData, VideoMeta)
rules/ -- Remotion teaching content served by rule tools
resources/remotion-player/ -- Widget source (React + Remotion Player)
widget.tsx -- Main widget component
components/
CodeComposition.tsx -- Bundle compiler (eval + runtime shim)
Tool schema
create_video({
files: string, // REQUIRED -- JSON string of {path: code}
entryFile?: string, // Default: "/src/Video.tsx"
title?: string,
durationInFrames?: number, // Default: 150
fps?: number, // Default: 30
width?: number, // Default: 1920
height?: number, // Default: 1080
})
The files parameter is a JSON string mapping virtual file paths to source code:
{
"/src/Video.tsx": "import {AbsoluteFill} from \"remotion\";\nexport default function Video() { return <AbsoluteFill />; }"
}
Session behavior
- Each MCP session maintains its own project state
- Calling
create_videomerges new files with the previous project - Metadata (title, fps, dimensions) carries forward unless overridden
- Sessions are capped at 250 concurrent projects with LRU eviction
License
MIT -- see LICENSE for details.
Note: Remotion is a dependency with its own license. Free for individuals and companies with up to 3 employees. Larger organizations need a company license.
Built with mcp-use and Remotion. Rule tools adapted from the remotion-best-practices skill by Remotion.
推荐服务器
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。
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。
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。
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。
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。
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。
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。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
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。

