YouTube Notes Fetcher

An Obsidian plugin that fetches YouTube video transcripts and auto-generates structured notes with AI summarization.

English | 中文

---

English

Features

• 📥 Single video fetch — Paste a YouTube URL, get a complete note
• 📺 Full channel archive — Pull every historical video from a creator (auto-skip existing)
• ⚡ Smart incremental updates — Stops on first known video; typical run completes in 5–10 seconds
• 🔔 Multi-channel subscriptions — Save and manage multiple channels
• 🤖 AI summarization — Built-in support for Claude API, OpenAI API, and Claude Code CLI (no API key needed)
• ⏰ Scheduled auto-update — Background polling for new videos
• 🎨 Customizable templates — Full control over note format
• 🖱️ Context menu integration — Right-click any YouTube URL in editor to import
• 🌐 Proxy support — Built-in proxy config for restricted regions

Three fetch modes

Mode	Use case	Speed
Latest N	First-time subscription or routine pull	Fast
Check updates (recommended for scheduled)	Stops at first known video	5–10s typical
Full history	First archive of a new creator	Minutes to tens of minutes

Architecture

This plugin uses a JS frontend + Python backend split:

• Frontend (main.js) — Obsidian UI, settings, command registration
• Backend (backend.py) — yt-dlp / youtube-transcript-api / AI calls

Installation

Prerequisites

pip install yt-dlp youtube-transcript-api

# Optional: API-based AI providers
pip install anthropic   # for Claude API
pip install openai      # for OpenAI API

If you use Claude Code CLI as the AI provider, no Python AI package is needed — just have claude available in PATH.

Manual install

1. Download main.js, manifest.json, styles.css, backend.py from the latest release
2. Place them under <your-vault>/.obsidian/plugins/youtube-notes-fetcher/
3. Reload Obsidian, then enable the plugin in Settings → Community plugins

Configuration

Open the plugin's settings tab:

Setting	Description
Python path	python by default; set full path if multiple versions exist
Proxy	e.g. http://127.0.0.1:10809 for V2Ray, http://127.0.0.1:7890 for Clash
Output directory	Relative to vault root
Max videos per fetch	Cap on batch fetch count
AI provider	claude_cli (no key) / claude (API) / openai (API)
AI API key	Required for claude and openai providers

Usage

Command palette (Ctrl+P / Cmd+P)

• Fetch single YouTube video
• Fetch latest from subscribed channels
• Manage subscribed channels
• Test connection

Right-click menu

Select a YouTube URL in any note → right-click → Import as YouTube note

Ribbon icon

Click the YouTube icon in the left sidebar.

Template variables

Customize the note template in settings using these placeholders:

Variable	Description
{{title}}	Video title
{{url}}	Video URL
{{author}}	Channel / author name
{{upload_date}}	Upload date
{{duration}}	Length
{{view_count}}	View count
{{description}}	Video description
{{transcript}}	Full transcript with [mm:ss] timestamps
{{ai_summary}}	AI-generated structured summary
{{created}}	Fetch date (YYYY-MM-DD)
{{fetched_at}}	Fetch timestamp

Privacy & data

• All processing is local. No data leaves your machine except:
  • YouTube API calls (via yt-dlp) to fetch metadata/transcripts
  • AI provider calls (Anthropic / OpenAI / Claude CLI) when summarization is enabled
• Your data.json (settings + subscriptions) and any cookies.txt (if you use one) stay in your vault and are listed in .gitignore.
• The plugin never sends API keys, cookies, or note content anywhere except the configured AI provider when you opt in.

Troubleshooting

"Missing dependency" → run pip install yt-dlp youtube-transcript-api

Network timeout → set a proxy. China users typically need this.

Python not found → set full path in settings, e.g. C:\Python313\python.exe

Transcript unavailable → some videos have no captions. Use the Test connection command to verify your environment.

License

MIT — see LICENSE

Support / 支持作者

If this plugin helps you, consider sponsoring development — WeChat / Afdian (爱发电) / GitHub Sponsors all welcome.

如果觉得有用，欢迎到 SPONSORS.md 通过微信赞赏 / 爱发电 / GitHub Sponsors 支持作者。

Contributing

Issues and PRs welcome. Please don't include your real data.json, cookies.txt, or any personal data when reporting bugs.

---

中文

功能特性

• 📥 单视频抓取：粘贴 YouTube URL，一键生成笔记
• 📺 频道全量归档：抓取博主所有历史视频（自动跳过已存在）
• ⚡ 智能增量更新：定期检查时遇到旧视频立即停止，5–10 秒完成
• 🔔 多频道订阅：保存并管理多个频道
• 🤖 AI 自动总结：内置 Claude API / OpenAI API / Claude Code CLI（免 key）三种 provider
• ⏰ 定时自动更新：后台定期检查新视频
• 🎨 自定义模板：完全控制笔记格式
• 🖱️ 右键菜单：选中编辑器里的 YouTube 链接右键导入
• 🌐 代理支持：内置代理配置

安装

pip install yt-dlp youtube-transcript-api

# 可选（用 API 时才需要）
pip install anthropic   # Claude API
pip install openai      # OpenAI API

如果使用 Claude Code CLI，无需安装 Python AI 包，只要 claude 命令在 PATH 里即可。

下载 release 中的 main.js、manifest.json、styles.css、backend.py，放到 <你的vault>/.obsidian/plugins/youtube-notes-fetcher/，然后在 Obsidian 设置 → 第三方插件中启用。

隐私

• 所有处理都在本地完成。除以下情况外不会向外发送数据：
  • yt-dlp 调用 YouTube 获取元数据和字幕
  • 启用 AI 摘要时调用对应 provider（Anthropic / OpenAI / Claude CLI）
• 你的 data.json（含订阅、API key）和 cookies.txt（如有）都不会进入仓库，已加入 .gitignore

故障排除

现象	解决
缺少依赖	pip install yt-dlp youtube-transcript-api
网络超时	配置代理。Clash 默认 7890，V2Ray 默认 10809
Python 找不到	设置里填完整路径
字幕获取失败	部分视频无字幕；用「连接测试」检查环境

License

MIT