yt-dlp-mcp-server
Enables AI agents to inspect, plan, download, and postprocess media from YouTube and other sites using yt-dlp, with safety and full option support.
README
yt-dlp-mcp-server
<div align="center">
The production-grade MCP server for yt-dlp — inspect, plan, download, postprocess media URLs for AI agents
28 typed tools. YouTube/youtu.be aware. Plan before write. Safe execution. Full yt-dlp option truth.
MCP server pre-release: v0.1.1 is ready for early users, but pin versions in production while the public release process stabilizes.
Quick Start • Tools • Usage Examples • Safety Model • Docker
</div>
Why This One
Most yt-dlp MCP servers are thin wrappers around 5–10 commands. This is a control plane for agents:
| Other servers | This server | |
|---|---|---|
| Tool count | ~10 | 28 typed tools |
| Plan before write | No | ytdlp_plan_download explains paths, deps, risks |
| Full option truth | Hardcoded flags | 323 options from upstream yt_dlp.options.create_parser() |
| Postprocessing | Basic audio extract | Remux, recode, embed assets, split/remove chapters, SponsorBlock |
| Archive management | None | Inspect archive entries, check exact entries, plan archive-safe downloads |
| Safe execution | Shell strings | argv arrays, shell: false, managed output roots, redaction |
| Raw escape hatch | Always on | Expert mode, gated behind env flag |
| Docker | None | Bundled yt-dlp, ffmpeg, ffprobe, Deno |
Quick Start
Step 1 — Install media tools
| OS | Commands |
|---|---|
| Windows | winget install -e --id OpenJS.NodeJS.LTS<br>winget install -e --id yt-dlp.yt-dlp |
| macOS | brew install node yt-dlp ffmpeg |
| Ubuntu/Debian | sudo apt install -y nodejs ffmpeg pipx && pipx install yt-dlp |
Windows:
yt-dlp.yt-dlpinstalls Deno andyt-dlp.FFmpegas winget dependencies. Restart PowerShell after install so the new command aliases are onPATH.
Step 2 — Add to your MCP client
{
"mcpServers": {
"yt-dlp": {
"command": "npx",
"args": ["-y", "yt-dlp-mcp-server@latest"],
"env": {
"YTDLP_MCP_OUTPUT_ROOT": "/Users/YOUR_USER/Downloads/yt-dlp-mcp"
}
}
}
}
Windows: use
"command": "npx.cmd"instead of"npx".
For Codex Desktop on Windows, use the TOML example in Client Setup instead of this JSON shape.
Step 3 — Verify
Ask your AI assistant: "Check the yt-dlp environment" — it will call ytdlp_check_environment and report which tools are found.
Client Setup
<details open> <summary><strong>Claude Desktop / Cursor / Windsurf / Cline</strong></summary>
Use the config from Step 2 above.
Config file locations:
- Claude Desktop macOS:
~/Library/Application Support/Claude/claude_desktop_config.json - Claude Desktop Windows:
%APPDATA%\Claude\claude_desktop_config.json - Cursor:
.cursor/mcp.jsonin a project, or global Cursor MCP settings - Windsurf:
~/.codeium/windsurf/mcp_config.json - Cline: extension MCP settings
</details>
<details> <summary><strong>Claude Code</strong></summary>
claude mcp add-json yt-dlp '{"type":"stdio","command":"npx","args":["-y","yt-dlp-mcp-server@latest"],"env":{"YTDLP_MCP_OUTPUT_ROOT":"~/Downloads/yt-dlp-mcp"}}'
</details>
<details> <summary><strong>VS Code</strong></summary>
VS Code requires a servers root and type field:
{
"servers": {
"yt-dlp": {
"type": "stdio",
"command": "npx",
"args": ["-y", "yt-dlp-mcp-server@latest"],
"env": {
"YTDLP_MCP_OUTPUT_ROOT": "/Users/YOUR_USER/Downloads/yt-dlp-mcp"
}
}
}
}
</details>
<details> <summary><strong>Warp</strong></summary>
mcp_servers:
yt-dlp:
command: npx
args:
- -y
- yt-dlp-mcp-server@latest
env:
YTDLP_MCP_OUTPUT_ROOT: /Users/YOUR_USER/Downloads/yt-dlp-mcp
</details>
<details> <summary><strong>Codex</strong></summary>
Codex Desktop usually reads config.toml from:
- Windows:
C:\Users\YOUR_USER\.codex\config.toml - macOS/Linux:
~/.codex/config.toml
[mcp_servers.yt-dlp]
enabled = true
command = "npx"
args = ["-y", "yt-dlp-mcp-server@latest"]
[mcp_servers.yt-dlp.env]
YTDLP_MCP_OUTPUT_ROOT = "/Users/YOUR_USER/Downloads/yt-dlp-mcp"
Windows:
[mcp_servers.yt-dlp]
enabled = true
command = "npx.cmd"
args = ["-y", "yt-dlp-mcp-server@latest"]
[mcp_servers.yt-dlp.env]
YTDLP_MCP_OUTPUT_ROOT = 'C:\Users\YOUR_USER\Downloads\yt-dlp-mcp'
</details>
Docker
Docker bundles everything — yt-dlp, ffmpeg, ffprobe, Deno. No host install required.
docker build -t yt-dlp-mcp-server:local .
mkdir -p "$HOME/Downloads/yt-dlp-mcp"
{
"mcpServers": {
"yt-dlp": {
"command": "docker",
"args": [
"run", "-i", "--rm",
"-v", "/Users/YOUR_USER/Downloads/yt-dlp-mcp:/downloads",
"-e", "YTDLP_MCP_OUTPUT_ROOT=/downloads",
"yt-dlp-mcp-server:local"
]
}
}
}
<details> <summary>Windows Docker config</summary>
{
"mcpServers": {
"yt-dlp": {
"command": "docker.exe",
"args": [
"run", "-i", "--rm",
"-v", "C:\\Users\\YOUR_USER\\Downloads\\yt-dlp-mcp:/downloads",
"-e", "YTDLP_MCP_OUTPUT_ROOT=/downloads",
"yt-dlp-mcp-server:local"
]
}
}
}
</details>
Tools
Environment
| Tool | What it does |
|---|---|
ytdlp_check_environment |
Check host readiness for yt-dlp, ffmpeg, ffprobe, PATH, cookies, plugins, and policy |
ytdlp_list_extractors |
List all supported site extractors from the installed yt-dlp binary |
ytdlp_list_impersonation_targets |
List browser impersonation targets for extractor troubleshooting |
Inspect (read-only, no file writes)
| Tool | What it does |
|---|---|
ytdlp_search_videos |
Search YouTube without downloading and return compact video results |
ytdlp_get_metadata |
Inspect title, duration, uploader, formats, subtitles, thumbnails, and selected best format |
ytdlp_list_formats |
List available quality, codec, resolution, bitrate, and format IDs |
ytdlp_list_subtitles |
List manual subtitle and automatic caption languages and formats |
ytdlp_list_thumbnails |
List thumbnail URLs, resolutions, and poster options |
ytdlp_probe_url |
Check URL support and identify the yt-dlp extractor without downloading |
Agent routing
When the user asks to download or save a media URL, including youtube.com and youtu.be links, use the download tools rather than trying to download in the model. A request for "best quality" can use the default video format selection, which plans as bestvideo*+bestaudio/best unless the user asks for a different format.
Plan (dry-run, explains before writing)
| Tool | What it does |
|---|---|
ytdlp_plan_download |
Dry-run a download and show output paths, best-quality format, dependencies, risks, and argv |
ytdlp_plan_postprocess |
Dry-run remux, recode, audio extraction, embedding, chapter, and SponsorBlock workflows |
ytdlp_validate_options |
Validate typed yt-dlp options and inspect the source-derived option catalog |
Download
| Tool | What it does |
|---|---|
ytdlp_download_video |
Download a web video URL, including YouTube and youtu.be links, with best-quality/default format support |
ytdlp_download_audio |
Download or extract audio-only output such as MP3, M4A, Opus, WAV, best audio, or podcast/music extraction |
ytdlp_download_subtitles |
Download subtitles or captions, including timestamped transcript files, without downloading the video |
ytdlp_download_thumbnail |
Download thumbnail, cover image, poster frame, or all thumbnails |
ytdlp_download_playlist |
Download playlist/channel entries with ranges, archive skipping, and max-download controls |
Archive
| Tool | What it does |
|---|---|
ytdlp_inspect_archive |
List saved entries in a yt-dlp download archive file |
ytdlp_check_archive |
Check if an exact yt-dlp archive entry is already present |
ytdlp_update_archive |
Plan archive handling and recommend download-archive usage; does not write the archive |
Download + postprocess
| Tool | What it does |
|---|---|
ytdlp_remux |
Download a URL and remux to a different container without re-encoding |
ytdlp_recode |
Download a URL and re-encode video to a target format |
ytdlp_extract_audio |
Download a URL and extract audio as mp3, m4a, opus, etc. |
ytdlp_embed_assets |
Download a URL and embed subtitles, thumbnails, metadata, or chapters |
ytdlp_split_chapters |
Download a URL and split media into one file per chapter |
ytdlp_remove_chapters |
Download a URL and remove chapters matching regex filters |
ytdlp_apply_sponsorblock |
Download a URL and remove or mark SponsorBlock segments |
Expert
| Tool | What it does |
|---|---|
ytdlp_execute_expert |
Last-resort raw yt-dlp argv escape hatch; requires YTDLP_MCP_ENABLE_EXPERT=true |
Usage Examples
Search and discover
"Search for Python tutorials on YouTube"
"Find the top 10 machine learning videos from this week"
"What formats are available for this video?"
"List all subtitle languages for https://youtube.com/watch?v=..."
Inspect before committing
"What metadata does this video have?"
"Check if this URL is supported"
"List all available formats with their codecs and bitrates"
"Plan a 720p download of this video — what will it create?"
Download
"Download this video in 720p: https://youtube.com/watch?v=..."
"Download just the audio as MP3"
"Download English auto-generated subtitles"
"Download this video from 1:30 to 3:00 only"
"Download playlist items 1–10 and skip ones I've already downloaded"
Postprocess
"Extract MP3 audio from this URL"
"Download this URL and remux it to MP4 without re-encoding"
"Download this URL and embed subtitles and thumbnail"
"Download this URL and split it into chapters"
"Download this URL and remove sponsor segments using SponsorBlock"
Archive management
"Check if this exact archive entry is already in my archive"
"Show me everything in my download archive"
"Plan how to archive-protect this playlist download"
Safety Model
| Policy | Behavior |
|---|---|
| Process execution | spawn with argv arrays, shell: false — no shell injection possible |
| Planning | Download and postprocess tools call the planner before writing any file |
| Output roots | All output paths constrained under managed roots by default |
| Archives | Archive paths resolve under output root unless arbitrary paths are enabled |
| Config files | Cookie/browser/netrc sources require YTDLP_MCP_ALLOW_CONFIG_FILES=true or server-level env |
| Raw args | Normal tools take typed inputs only; expert mode is gated |
| Redaction | Cookies, PO tokens, auth headers, signed URL params redacted in previews and errors |
Environment Variables
| Variable | Default | Purpose |
|---|---|---|
YTDLP_MCP_OUTPUT_ROOT |
~/Downloads/yt-dlp-mcp |
Managed output directory |
YTDLP_MCP_TEMP_ROOT |
OS temp + yt-dlp-mcp |
Managed temp directory |
YTDLP_MCP_YTDLP_PATH |
auto-detected | Override yt-dlp binary path |
YTDLP_MCP_FFMPEG_PATH |
auto-detected | Override ffmpeg binary path |
YTDLP_MCP_FFPROBE_PATH |
auto-detected | Override ffprobe binary path |
YTDLP_MCP_ENABLE_EXPERT |
false |
Enable ytdlp_execute_expert |
YTDLP_MCP_ALLOW_ARBITRARY_OUTPUT_PATHS |
false |
Allow paths outside managed roots |
YTDLP_MCP_ALLOW_CONFIG_FILES |
false |
Allow cookie/netrc/config file inputs |
YTDLP_MCP_COOKIES_FILE |
— | Server-approved cookies file |
YTDLP_MCP_COOKIES_FROM_BROWSER |
— | Server-approved browser cookie source |
YTDLP_MCP_TIMEOUT_MS |
900000 (15 min) |
Per-command timeout |
YTDLP_MCP_MAX_OUTPUT_BYTES |
4194304 (4 MiB) |
Max retained stdout/stderr |
If doctor reports missing binaries that work in your terminal, set explicit paths:
{
"YTDLP_MCP_YTDLP_PATH": "/opt/homebrew/bin/yt-dlp",
"YTDLP_MCP_FFMPEG_PATH": "/opt/homebrew/bin/ffmpeg",
"YTDLP_MCP_FFPROBE_PATH": "/opt/homebrew/bin/ffprobe"
}
Use /usr/local/bin/... for Intel Homebrew on macOS. Use absolute .exe paths on Windows.
CLI
The yt-dlp-mcp-server binary is useful for humans and quiet when launched by an MCP client.
yt-dlp-mcp-server doctor # check environment
yt-dlp-mcp-server doctor --json # machine-readable
yt-dlp-mcp-server print-config --client claude --mode npx
yt-dlp-mcp-server print-config --client codex --mode npx --os windows
yt-dlp-mcp-server print-config --client all --mode docker
yt-dlp-mcp-server print-deps --os macos
yt-dlp-mcp-server print-deps --os windows --manager winget
yt-dlp-mcp-server print-deps --os linux --manager apt
yt-dlp-mcp-server stdio # start MCP server explicitly
Supported --client values: claude, cursor, windsurf, cline, vscode, warp, codex, all.
Resources and Prompts
The server exposes MCP resources and prompt templates:
| Type | Name | Purpose |
|---|---|---|
| Resource | ytdlp://capabilities |
Tool inventory and capability summary |
| Resource | ytdlp://option-catalog/groups |
Full 323-option catalog by group |
| Resource | ytdlp://safety-policy |
Current safety policy settings |
| Resource | ytdlp://troubleshooting |
Common errors and fixes |
| Resource | ytdlp://environment |
Live environment check snapshot |
| Prompt | archive_playlist_safely |
Guided workflow for archive-protected playlist downloads |
| Prompt | choose_smallest_acceptable_format |
Guided format selection for minimal file size |
Development
pnpm install
pnpm run validate:ci # typecheck + tests + build + MCP contract + output audit
pnpm run validate:local # above + live smoke tests (requires host media tools)
pnpm run validate:release # above + Docker smoke tests
MCP Inspector for manual UX check:
npx -y @modelcontextprotocol/inspector node dist/index.js
License
MIT — see 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。