youtube-mcp
Enables AI agents to fetch transcripts, metadata, and download videos/audio from YouTube without API keys.
README
YouTube MCP server and CLI
Zero Setup · No API Key · No Account · Open Source
YouTube is full of knowledge, but it is trapped there. AI agents cannot access it natively. This repo ships two packages that fix that: an MCP server for AI agents and a CLI for the terminal.
- Fetch transcript (plain text or with timestamps)
- Download transcript as
.md - Get video metadata
- Search within captions
- Download video or audio
Available as:
All locally. No API keys required.
Table of Contents
MCP Server
Tools
| Tool | Description |
|---|---|
get_transcript |
Fetches the full transcript as plain text |
get_transcript_timed |
Transcript with [MM:SS] timestamps per segment |
download_transcript |
Saves transcript as a .md file |
download_transcript_timed |
Saves transcript as a .md file with timestamps |
get_metadata |
Returns title, channel, publish date, view count, duration, likes, and description |
search_transcript |
Searches for a keyword or phrase and returns matching segments with timestamps |
download_video |
Downloads video+audio as .mp4 |
download_audio |
Downloads audio as .mp3 (or m4a, flac, opus, ...) |
download_video and download_audio are fully self-contained. yt-dlp and ffmpeg are downloaded automatically on npm install.
Installation
Method A: let your AI do it
Tell your AI agent:
Add youtube-mcp as an MCP server. Run it with: npx @umbertotancorre/youtube-mcp
Method B: manual config
Add this to your MCP client config (e.g. claude_desktop_config.json):
{
"mcpServers": {
"youtube": {
"command": "npx",
"args": ["-y", "@umbertotancorre/youtube-mcp"]
}
}
}
All tools appear automatically. No install needed, npx handles it on first run.
CLI
Commands
youtube-cli transcript <url> Print transcript as plain text
youtube-cli transcript <url> --timestamps Include [MM:SS] timestamps
youtube-cli transcript <url> --save Save as .md to Downloads
youtube-cli transcript <url> --language it Fetch in a specific language
youtube-cli search <url> <query> Search transcript with timestamps
youtube-cli metadata <url> Print title, channel, views, duration
youtube-cli metadata <url> --json Output as JSON
youtube-cli download <url> Download video to Downloads
youtube-cli download <url> --quality hd1080 Choose quality
youtube-cli download <url> --audio Download audio only (mp3)
youtube-cli download <url> --audio --format flac
youtube-cli completions Print bash completion script
youtube-cli completions zsh Print zsh completion script
Downloads always go to the platform-native Downloads folder:
| OS | Path |
|---|---|
| macOS | ~/Downloads |
| Linux | $XDG_DOWNLOAD_DIR or ~/Downloads |
| Windows | %USERPROFILE%\Downloads |
Installation
Global install (recommended):
npm install -g @umbertotancorre/youtube-cli
youtube-cli --help
One-off use without installing:
npx @umbertotancorre/youtube-cli transcript dQw4w9WgXcQ
Shell completions (optional):
# bash
source <(youtube-cli completions)
# zsh
youtube-cli completions zsh > ~/.zsh/completions/_youtube-cli
Development
This is an npm workspaces monorepo with three packages:
youtube-mcp-cli/
packages/
core/ # shared logic (transcripts, metadata, downloads)
mcp/ # MCP server (@umbertotancorre/youtube-mcp)
cli/ # CLI tool (@umbertotancorre/youtube-cli)
npm run build # build all packages (core, mcp, cli in order)
npm run dev:mcp # build core, then run MCP server via ts-node
npm run dev:cli # build core, then run CLI via ts-node
Disclaimer
This project only accesses publicly available YouTube data: the same captions, metadata, and streams visible in any browser. It does not bypass authentication, paywalls, or age gates. No API key, account, or login is required.
End users are responsible for complying with YouTube's Terms of Service. The maintainer does not host or operate any service. This is a local tool you run on your own machine.
License
youtube-mcp-cli is fully open source, licensed under the MIT License.
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。