The enhanced transcript tool is what separates it: several videos at once, filtered to a time range or a keyword, with timestamps and metadata — which turns "what did these five talks say about X" into one call instead of five transcript reads.
A fuller YouTube toolkit built on the Data API: searching with real filters, channel and video detail, comments with sorting, related and trending video lists, plus a transcript layer that goes beyond fetching text — filtering, searching and multi-video extraction.
- Videos — `search_videos` with filtering by channel, duration and region, `get_video_details` for title, channel, views and likes, `get_video_comments` with sorting, `get_related_videos`, `get_trending_videos` by region
- Channels — `get_channel_details` for name, subscribers and views
- Transcripts — `get_video_transcript` in the languages you name, and `get_video_enhanced_transcript` for filtering, keyword search, time-bounded slices, timestamped output and several videos at once
- `transcript_summary` — a prompt that summarises a video from its transcript with options you set
- Resources addressed by URI: `youtube://video/{video_id}`, `youtube://channel/{channel_id}`, `youtube://transcript/{video_id}?language={language}`, and `youtube://available-youtube-tools` for the tool list itself
Python 3.12 or higher and a YouTube API key: create a Google Cloud project, enable the YouTube Data API v3, create an API key, and put it in `YOUTUBE_API_KEY`. Install with uv into a 3.12 virtual environment, or build the Docker image and pass the key as an environment variable; Smithery installs it into Claude Desktop in one command. Because search, comments and trending all go through the Data API, they draw on that key's quota — the transcript tools are the ones that keep working when it runs out.
One command plus a key — npx -y @smithery/cli install @jikime/py-mcp-youtube-toolbox --client claude, then supply credentials