Playlist → RAG corpus

What you get

• Chunks that respect caption boundaries instead of cutting sentences
• Metadata per chunk: video id, title, channel, start second
• Citations that open the video at the right moment (&t=123s)

With an AI assistant

Connect the MCP server to Claude, ChatGPT or Cursor, then paste:

With the API

Set GYT_API_KEY to a key from the developers page. Each transcript or list page costs one credit; checking a channel's latest uploads is free.

1. 1. List every video (follow continuation until it is null)

   curl -s -H "Authorization: Bearer $GYT_API_KEY" \
     "https://getyoutubetranscript.com/api/v1/playlist?list=PLAYLIST_ID"

2. 2. Fetch timed segments for each video (they are in data.segments)

   curl -s -H "Authorization: Bearer $GYT_API_KEY" \
     "https://getyoutubetranscript.com/api/v1/transcript?v=VIDEO_ID&timestamps=true"

3. 3. Chunk on segment boundaries (Python)

   def chunks(video, segments, max_chars=1200):
       buf, start = [], None
       for s in segments:
           start = s["start"] if start is None else start
           buf.append(s["text"])
           if sum(len(t) for t in buf) > max_chars:
               yield {"text": " ".join(buf), "video_id": video, "start": start,
                      "url": f"https://youtu.be/{video}?t={int(start)}"}
               buf, start = [], None
       if buf:
           yield {"text": " ".join(buf), "video_id": video, "start": start,
                  "url": f"https://youtu.be/{video}?t={int(start)}"}

Limits to know

• Transcripts cached before segments were stored come back without timings; chunk those by characters.
• Re-fetching a transcript costs a credit each time, so store what you ingest.

Tools for this workflow