take_notes

One URL in, formatted notes out.

How It Works

1. Downloads audio from the URL (YouTube, Vimeo, TikTok, etc.)
2. Transcribes the audio locally using faster-whisper
3. Saves a .md file to your Desktop
4. Returns the full transcription so Claude can rewrite the file with clean, formatted notes

The output is always .md with YAML frontmatter. Notes are formatted for Obsidian-style readability.

---

Styles

Note formatting ranges from minimal to maximum visual richness:

Style	What you get
tldr	Shortest summary. Bold key terms, flat bullets, one screen max.
notes	Clean sections + nested bullets. The default.
highlight	Notes with callout blocks for key insights, blockquotes with timestamps.
eye-candy	Maximum visual formatting: callouts, tables, checklists, blockquotes. The full Obsidian treatment.
quiz	Multiple-choice questions generated from the content, answer key at bottom.

You don’t need to specify a style. Claude picks one based on what you ask for. Or be explicit:

“Take eye-candy notes from https://youtube.com/watch?v=xxx”

“Give me a quiz from this lecture: https://youtube.com/watch?v=xxx”

---

Example

Request:

{
  "url": "https://youtube.com/watch?v=xxx",
  "style": "highlight"
}

Response:

{
  "success": true,
  "md_path": "/Users/you/Desktop/Video Title.md",
  "audio_path": "/Users/you/Downloads/Video Title.webm",
  "title": "Video Title",
  "duration": 1662.0,
  "duration_formatted": "27:42",
  "language": "en",
  "style": "highlight",
  "transcription": "Full transcription text...",
  "instruction": "..."
}

After receiving the response, Claude rewrites the .md with formatted notes matching the selected style.

---

Parameters

Parameter	Required	Default	Description
url	No	—	Video/audio URL (YouTube, Vimeo, TikTok, Twitter, SoundCloud, etc.). Required for the first call.
save_content	No	—	Formatted notes to save. When provided, writes this content to the file from the previous take_notes call. No url needed.
output_path	No	—	Explicit file path to save notes to. Use when saving notes from a memory transcript (no prior take_notes URL call).
style	No	notes	Note style: tldr, notes, highlight, eye-candy, quiz
output_dir	No	~/Desktop	Directory to save the .md file
model_size	No	tiny	Whisper model size
read_aloud	No	false	Generate a spoken MP3 summary and embed an audio player in the notes for Obsidian playback
visual	No	—	Query for visual context extraction. Downloads the video, finds transcript moments matching the query, extracts screenshots, and embeds them inline in the notes with ![[filename.png]] Obsidian syntax.

---

Notes

Transcriptions are stored in memory normally. Re-running on the same URL is instant.

The audio file is saved to ~/Downloads. The .md notes file is saved to ~/Desktop.

Works with any URL that yt-dlp supports, including 1000+ sites.

Multilingual: Non-English audio is transcribed in its original language. Claude translates while formatting notes. After saving, a translation offer lets you store an English (eng) version in memory alongside the original.

---

Obsidian Setup

Every .md on your Mac opens in Obsidian with real-time sync