Your AI editor is great at code and useless at video. Ask Claude or Cursor about a 40-minute talk and it shrugs — it can't watch YouTube. So you do the dance: open the video, hit "show transcript," copy the wall of text, strip the timestamps, paste it in. Every. Single. Time.
MCP kills that loop — you give your agent a tool it can call, and it fetches the transcript itself, mid-conversation.
The setup: drop Scribefy into your MCP config (npx -y scribefy-mcp + SCRIBEFY_API_KEY), grab a key at scribefy.app, restart → tools appear. (links to the Cursor / Windsurf / Claude guides)
What your agent gets: extract_transcript, search_videos, get_video_metadata, get_related_videos.
3 ways I use it: summarize a talk with timestamps · lift commands out of a coding tutorial · search + compare across multiple videos in one prompt.
Why a tool beats a script: the model decides when to fetch, which video, which language — "I have my answer," not "I have a transcript."
https://scribefy.app/ - Free credits for signed in users
Top comments (0)