Mmk Youtube Channel Summary

Recipe: YouTube Channel Summary to Notion

Prerequisite: Read ../mmk-shared/SKILL.md for auth, global flags, and error handling. Uses: mmk-youtube-videotype — Filter out Shorts. Uses: mmk-youtube-metadata — Video metadata (views, likes, etc.). Uses: mmk-youtube-transcript — Video transcript for summarization. Uses: mmk-notion-database — Schema discovery and upsert commands.

A fully automatic workflow that fetches recent videos from one or more YouTube channel RSS feeds, filters out Shorts, collects metadata and transcripts, generates AI summaries, and upserts everything into a Notion database.

This recipe has side effects (creates/updates records in a Notion database). It requires manual invocation only.

No confirmation prompts — the recipe runs end-to-end automatically once invoked with the required parameters.

Database Schema Requirements

The recipe discovers properties by their Notion type and semantic meaning, not by exact name. English and Korean property names are both supported automatically.

Required Properties

Optional Properties (auto-detected for enrichment)

Notes

• Title property as VideoID: The title property stores the video ID (e.g., dQw4w9WgXcQ) and serves as the upsert lookup key. This prevents duplicates on re-runs. The human-readable video title goes into a separate rich_text property.
• Type-first matching: The title type is unique per database (always unambiguous). For url, number, and rich_text types, name patterns resolve ambiguity.
• Ambiguity handling: If multiple properties match a required purpose, present options and ask the user to choose.
• Summary property: If no Summary property is found, AI summaries will be displayed in the final report but not written to Notion.

Workflow

This workflow runs fully automatically. For multiple channel IDs, Steps 1-6 repeat per channel. Steps 7-9 run once after all channels are processed.

Step 1: Fetch RSS feed

curl -s "https://www.youtube.com/feeds/videos.xml?channel_id=CHANNEL_ID"

Parse the XML response to extract <entry> elements. From each entry, collect:

• <yt:videoId> — video ID
• <title> — video title
• <published> — publish date (ISO 8601)
• <link href="..."/> — video URL
• <media:group><media:thumbnail url="..." /></media:group> — thumbnail URL

If the feed returns empty or an error, log the failure and skip to the next channel.

Step 2: Filter by date

Filter entries to only those published within the last N days (default: 7).

If no entries remain, log "No videos published in the last N days for channel [ID]" and skip to the next channel.

Step 3: Filter out Shorts

For each remaining video, check the video type:

mmk youtube videotype <video_id> -o json

Keep only entries where type is "video". Discard entries where type is "short".

If no videos remain, log "All recent videos are Shorts for channel [ID]" and skip to the next channel.

Step 4: Get metadata

mmk youtube metadata <video_id> -o json

Collect from each response: title (authoritative), author/channel name, viewCount, likeCount, publishDate.

If metadata fails for a video, use the RSS title and channel name as fallback.

Step 5: Get transcripts

mmk youtube transcript <video_id> -o json

Collect the content field. If a video has no captions, mark as "No transcript available" and continue.

Step 6: AI-summarize

For each video:

1. If transcript available: generate a concise summary (2-4 sentences) covering the main topic and key points, plus 3-5 topic tags
2. If no transcript: generate a brief description from the title and metadata only, noting "(from metadata only)"

Step 7: Discover Notion database schema

mmk notion database schema --database-id <db-uuid> --enhanced -o json

7a. Required property discovery

Map database properties to required purposes using type-first, then name-pattern matching:

1. VideoID: Use the single title type property (always unambiguous).
2. Title: Find rich_text matching: Title, 제목, 영상제목.
3. Channel: Find rich_text matching: Channel, 채널, 채널명.
4. URL: Find url type properties. If exactly one, use it. If multiple, match by name: URL, 링크, 영상링크.
5. Published: Find date type properties. If exactly one, use it. If multiple, match by name: Published, 발행일, 게시일, 업로드일.

If any required purpose has zero candidates, stop with a clear error listing what is missing and what types/names are expected.

7b. Optional property discovery

Scan remaining properties for enrichment fields:

1. Thumbnail: url type matching Thumbnail, 썸네일
2. Summary: rich_text type matching Summary, 요약, AI요약
3. Topics: multi_select or rich_text type matching Topics, 주제, 태그, Tags
4. ViewCount: number type matching Views, 조회수
5. LikeCount: number type matching Likes, 좋아요, 좋아요수

Report discovered mappings and continue automatically.

Step 8: Upsert to Notion

Build the upsert data array using the discovered property names. Replace property keys with actual Notion property names from Step 7.

mmk notion database upsert --database-id <db-uuid> --lookup <title-property-name> --data '[
  {"<VideoID>":"abc123","<Title>":"How to Cook Pasta","<Channel>":"ChefChan","<URL>":"https://www.youtube.com/watch?v=abc123","<Published>":"2026-03-20","<Thumbnail>":"https://i.ytimg.com/vi/abc123/hqdefault.jpg","<Summary>":"A guide to making fresh pasta...","<ViewCount>":12345,"<LikeCount>":890},
  {"<VideoID>":"def456","<Title>":"Kitchen Tips","<Channel>":"ChefChan","<URL>":"https://www.youtube.com/watch?v=def456","<Published>":"2026-03-18","<Summary>":"Tips for organizing...","<ViewCount>":8901,"<LikeCount>":456}
]' -o json

Only include optional properties that were successfully discovered in Step 7b. For Topics with multi_select type, pass as an array (e.g., ["cooking","pasta"]). For rich_text type, pass as a comma-separated string.

Step 9: Present final summary

Display a single table with all results across all channels:

# YouTube Channel Summary Complete
- Channels processed: 3
- Period: Last 7 days
- Videos found: 15 | Shorts filtered: 4 | Regular videos: 11
- Transcripts: 9 available, 2 unavailable
- Upserted to Notion: 11 records (8 new, 3 updated)

| # | Channel  | Title              | Published  | Views  | Likes | Topics            | Summary                    |
|---|----------|--------------------|------------|--------|-------|-------------------|----------------------------|
| 1 | ChefChan | How to Cook Pasta  | 2026-03-20 | 12,345 | 890   | cooking, pasta    | A guide to making fresh... |
| 2 | ChefChan | Kitchen Tips       | 2026-03-18 | 8,901  | 456   | kitchen, tips     | Tips for organizing...     |
| 3 | DevTube  | Building CLI Tools | 2026-03-19 | 5,432  | 321   | CLI, Node.js      | Demonstrates building...   |

Parameters

Error Handling

Example Session

User: /mmk-youtube-channel-summary
      Channels: UC_x5XG1OV2P6uZZ5FSM9Ttw, UCddiUEpeqJcYeBxX1IVBKvQ
      Database: abc123-def456
      Last 7 days

Claude: [Channel 1/2] UC_x5XG1OV2P6uZZ5FSM9Ttw
       [Step 1] Fetching RSS feed... 15 entries found.
       [Step 2] Filtering to last 7 days... 10 videos.
       [Step 3] Checking video types... 8 regular, 2 Shorts filtered.
       [Step 4] Fetching metadata... 8/8 complete.
       [Step 5] Fetching transcripts... 7/8 available, 1 no captions.
       [Step 6] Generating summaries... 8/8 complete.

       [Channel 2/2] UCddiUEpeqJcYeBxX1IVBKvQ
       [Step 1] Fetching RSS feed... 12 entries found.
       [Step 2] Filtering to last 7 days... 6 videos.
       [Step 3] Checking video types... 5 regular, 1 Short filtered.
       [Step 4] Fetching metadata... 5/5 complete.
       [Step 5] Fetching transcripts... 5/5 available.
       [Step 6] Generating summaries... 5/5 complete.

       [Step 7] Discovering Notion database schema...

       Property mapping (auto-detected):
         VideoID   -> "VideoID" (title)
         Title     -> "제목" (rich_text)
         Channel   -> "채널" (rich_text)
         URL       -> "URL" (url)
         Published -> "게시일" (date)

       Optional fields discovered:
         Thumbnail  -> "썸네일" (url)
         Summary    -> "AI요약" (rich_text)
         Topics     -> "태그" (multi_select)
         ViewCount  -> "조회수" (number)
         LikeCount  -> "좋아요" (number)

       [Step 8] Upserting 13 videos to Notion...
       Upserted: 13 records (11 new, 2 updated).

       [Step 9] Complete!
       - Channels processed: 2
       - Period: Last 7 days
       - Videos: 16 found, 3 Shorts filtered, 13 regular
       - Transcripts: 12 available, 1 unavailable
       - Upserted: 13 records (11 new, 2 updated)

       | # | Channel    | Title                | Published  | Views  | Topics            | Summary                    |
       |---|------------|----------------------|------------|--------|-------------------|----------------------------|
       | 1 | ChefChan   | How to Cook Pasta    | 2026-03-20 | 12,345 | cooking, pasta    | A guide to making fresh... |
       | 2 | ChefChan   | Kitchen Organization | 2026-03-19 | 8,901  | kitchen, tips     | Tips for organizing...     |
       | 3 | DevTube    | Building CLI Tools   | 2026-03-19 | 5,432  | CLI, Node.js      | Demonstrates building...   |
       | ...                                                                                                       |

See Also

• mmk-shared — Auth, global flags, error handling
• mmk-youtube — All YouTube commands overview
• mmk-youtube-videotype — Video vs Short detection
• mmk-youtube-metadata — Video metadata
• mmk-youtube-transcript — Video transcript
• mmk-notion-database — Schema, query, upsert commands