youtube-mcp
MCP server for YouTube creator-ops — video metadata, comments, playlists, channel analytics, plus a ComfyUI bridge for AI thumbnail generation.
README
youtube-mcp
OAuth-authenticated YouTube MCP for channel owners — edit your video metadata, reply to and moderate comments, manage playlists, query channel analytics, and generate + set AI thumbnails via a ComfyUI bridge. Goes beyond the read-only Data API v3 wrappers that dominate this space.
Part of the MCP Server Series.
The pitch
Most existing YouTube MCPs use an API key against Data API v3 — search videos, fetch public metadata, read-only. This one uses OAuth 2.0 (Authorization Code + PKCE) so it can actually write to your channel: update video titles/descriptions/tags, reply to comments, moderate spam, manage playlists. Plus it hits the separate YouTube Analytics API for channel stats, and — the hero tool — generates a thumbnail via ComfyUI and pushes it to YouTube in a single MCP call.
Claude, use generate_and_set_thumbnail on video abc123:
prompt: "cyberpunk hacker at keyboard, neon blue and pink, high contrast"
→ ComfyUI renders 1280×720 → youtube-mcp fetches the bytes → POSTs to thumbnails.set. Done.
Install
# npx — no install
npx @miller-joe/youtube-mcp --help
# Docker
docker run -p 9120:9120 \
-e YOUTUBE_CLIENT_ID=... \
-e YOUTUBE_CLIENT_SECRET=... \
-e YOUTUBE_TOKEN_FILE=/token/token.json \
-v $PWD/token:/token \
ghcr.io/miller-joe/youtube-mcp:latest
Setup — Google Cloud one-time (~10 min)
-
Google account + YouTube channel — use a personal account, not a workspace one you might lose.
-
Google Cloud project at https://console.cloud.google.com — call it whatever (e.g.
youtube-mcp). -
Enable APIs:
- YouTube Data API v3
- YouTube Analytics API
-
OAuth consent screen — External, App name, support email; in Scopes add:
youtube.uploadyoutube.force-sslyt-analytics.readonly
-
Stay in Testing mode. Add yourself as a test user (required). As the project owner, your refresh token won't expire.
-
Create OAuth Client ID: Application type = Desktop app. Download the JSON.
-
Run the interactive auth flow:
npx @miller-joe/youtube-mcp --auth --client-secret-file ./client_secret.jsonA browser opens → you log in to the Google account tied to your YouTube channel → grant the requested scopes. On success, a refresh token is saved to
~/.config/youtube-mcp/token.json. -
Start the server:
npx @miller-joe/youtube-mcp --client-secret-file ./client_secret.jsonOr provide the client credentials via env:
YOUTUBE_CLIENT_SECRET_FILE, orYOUTUBE_CLIENT_ID+YOUTUBE_CLIENT_SECRET.
Connect an MCP client
claude mcp add --transport http youtube http://localhost:9120/mcp
Or point your MCP gateway at the Streamable HTTP endpoint.
Configuration
| CLI flag | Env var | Default | Notes |
|---|---|---|---|
--client-secret-file |
YOUTUBE_CLIENT_SECRET_FILE |
— | Path to Google OAuth JSON |
--client-id |
YOUTUBE_CLIENT_ID |
— | Alternative to the secret file |
--client-secret |
YOUTUBE_CLIENT_SECRET |
— | Alternative to the secret file |
--token-file |
YOUTUBE_TOKEN_FILE |
~/.config/youtube-mcp/token.json |
Refresh token storage |
--host |
MCP_HOST |
0.0.0.0 |
Bind host |
--port |
MCP_PORT |
9120 |
Bind port |
--comfyui-url |
COMFYUI_URL |
(unset — bridge disabled) | ComfyUI HTTP URL for bridge tools |
| — | COMFYUI_DEFAULT_CKPT |
sd_xl_base_1.0.safetensors |
Default checkpoint for bridge tool |
Tools
Videos
list_my_videos— paginated list of authenticated channel's uploadsget_video— full detail for one videoupdate_video_metadata— title / description / tags / category / privacydelete_video— permanently delete a video. Requiresconfirm_video_titleto match the current title exactly as a guard against deleting the wrong video.
Captions
list_captions— list caption tracks on a video (language, name, status, draft flag)upload_caption— upload an SRT or WebVTT caption track to a videodelete_caption— delete a caption track
Shorts
list_my_shorts— find Shorts in recent uploads (filters by ≤60s duration)get_shorts_analytics— YouTube Analytics query restricted to Shorts (creatorContentType==SHORTS)
Playlists
create_playlist— create a playlist (default private)add_to_playlist— add a video to an existing playlist
Comments
list_comments— top-level comment threads on a videoreply_to_comment— reply to a top-level commentmoderate_comment— hold / approve / reject a comment
Analytics
query_channel_analytics— date-ranged metrics with optional dimensions and filters
Bridge (when COMFYUI_URL is configured)
generate_and_set_thumbnail— generate thumbnail via ComfyUI and set it on a video in one call
Quota notes
YouTube Data API free tier = 10,000 units/day. Key operation costs:
videos.list,commentThreads.list— 1 unit eachvideos.update,comments.insert,thumbnails.set— 50 units eachvideos.insert(upload) — 1,600 units → ~6 uploads/day on free tier
Most creator-ops workflows stay well under the free cap.
Architecture
┌────────────────┐ ┌──────────────────┐ ┌─────────────────┐
│ MCP client │────▶│ youtube-mcp │────▶│ YouTube APIs │
│ (Claude etc.) │◀────│ (this server) │◀────│ (Data/Analytics)│
└────────────────┘ └────────┬─────────┘ └─────────────────┘
│
│ (bridge tools only)
▼
┌──────────────────┐
│ ComfyUI │
│ (txt2img) │
└──────────────────┘
OAuth refresh tokens are cached locally and refreshed just-in-time before expiry. The bridge tool downloads image bytes from ComfyUI internally — ComfyUI does not need to be publicly reachable.
Development
git clone https://github.com/miller-joe/youtube-mcp
cd youtube-mcp
npm install
npm run dev
npm run build
npm test
Requires Node 20+.
Roadmap
- [x] Video list / get / update metadata
- [x] Playlist create + add-to
- [x] Comments list / reply / moderate
- [x] Channel analytics query
- [x] ComfyUI thumbnail bridge (
generate_and_set_thumbnail) - [x] Caption upload + list + delete (
upload_caption,list_captions,delete_caption) - [x] Video delete with title-match confirm guard
- [x] Shorts ergonomics:
list_my_shorts(duration filter) +get_shorts_analytics(creatorContentType==SHORTS) - [ ] Video upload (
video_upload) — resumable upload support - [ ] Reporting API for bulk historical data exports
License
MIT © Joe Miller
Support
If this saves you time, consider supporting development:
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。
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。
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。
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。
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。
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。
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。
mcp-server-qdrant
这个仓库展示了如何为向量搜索引擎 Qdrant 创建一个 MCP (Managed Control Plane) 服务器的示例。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。