Call search(<natural language query>) to retrieve a list of YouTube video IDs related to the subject matter,
slightly filtered by the Agent.
Place a .env file under the natural-search dir with both SUBSCRIPTION_KEY (Unwrapathon) & YOUTUBE_API_KEY.
Run python db.py to init the db (under the db dir).
MCP is downloaded by running pip install "mcp[cli]"
... Weird, but necessary.
The higher the number of videos/comments you scrape, the more accurate the categorization will be. The lower the batch size you use for categorization, the more accurate the categorization will be. Both for a cost of course.