AI Use Case: Treat YouTube like a Knowledge Base
Build an AI Agent to transcribe videos and chat with them
Whenever I want to learn something new, YouTube is my go-to destination. This could be anything—from building a great product roadmap to discovering the latest science-based tools for living a healthy life. YouTube has also become the top platform for listening watching podcasts, another fantastic source of knowledge and insight. Capturing that knowledge in an actionable way used to require a lot of effort—manual note-taking and, ideally, distilling those notes later. Taking notes is a great way to remember things, but you can go further: by applying AI in a clever way, you can treat each YouTube video as a vast pool of knowledge you can actually query. You can summarize key concepts and practical frameworks, combine them with insights from related fields, and draft all sorts of documents based on that knowledge.
By combining n8n for workflow automation with Supabase as my vector database, I created an AI agent that can transcribe YouTube videos and answer questions about them. In fact, it’s three tools that I’m making available via MCP (Model Context Protocol) to any AI client that supports the protocol. MCP is an open standard created by Anthropic, the company behind Claude. It was quickly adopted and became a widely supported framework for providing context and tools to Large Language Models (LLMs).
This is the perfect project for anyone wanting to learn how to build AI agents. The general approach to storing data in a format that LLMs can use will always be the same. It works for any text-based content—for example, the notes from your notes app, a journal, or office documents. I personally find n8n a great foundation for all that work, but there are many alternatives that provide a similarly powerful no-code or low-code environment (e.g., Make, Zapier).
Solution Overview
Add YouTube video to knowledge base:
Retrieve knowledge from knowledge base:
Tools and services involved:
n8n: The platform for building automated workflows and transforming them into AI agents.
Supabase: Our vector store database platform.
Notion: Offers an overview of the YouTube videos added to the Knowledge Base (optional).
Claude: My preferred AI, with MCP support.
Google Drive: Used to store YouTube transcripts (optional).
Supadata: An easy-to-use third-party API for YouTube.
How it works
Instead of a lengthy step-by-step guide, I recorded a video that walks you through the entire workflow. The video covers:
Setting up the MCP Server trigger in n8n and connecting it to three tools—transcribe video, retrieve knowledge and overview of all videos in the knowledge base
Creating a new database page in Notion for each added YouTube video to build an index of your Knowledge Base (optional)
Retrieving the video transcript and details via the Supadata API
Storing the raw video transcript in Google Drive (optional)
Generating a polished video description with ChatGPT using the OpenAI API (optional)
Storing the video transcript in a vector database (Supabase) to make it accessible for an LLM
Alternative Solutions
If building a custom AI agent seems like overkill, here are some excellent tools for extracting knowledge from YouTube videos and podcasts:
NotebookLM: Google's research assistant has built-in YouTube support. Just paste your video links and start asking questions—completely free!
Snipd: This brilliant iPhone app reimagines the podcast player. As you listen, simply "snip" any memorable moments. The app automatically transcribes these segments, summarizes them, extracts key insights, and can export them to your preferred knowledge management system (Obsidian, Notion, Readwise, and others). And that's just scratching the surface.
Readwise Reader: A personal favorite that's been central to my knowledge management for years. Their Reader app handles YouTube videos beautifully—generating transcripts automatically and letting you highlight directly in the transcript as you watch.