pennydinh/penny-notebooklm-research

Penny NotebookLM Research automates a specific YouTube research workflow for content creators: find popular videos on a topic, read their actual transcripts, and turn that research into video ideas and a written script. Instead of an AI model guessing what a trend looks like, this tool has Google's NotebookLM and its Gemini model read the real transcript of each video it finds, so the analysis is grounded in what was actually said. It works as two connected pieces that both need to be installed together: an MCP server, which is a small local program that runs searches through Brave Search and Apify and controls NotebookLM on your behalf, and a Claude skill file that teaches Claude Cowork how to run the whole process as one coordinated flow. Neither piece works without the other. Setup starts with an install script that clones the project, installs its dependencies, builds it, and registers it with Claude, after which you add your own Brave Search API key, which is free for a limited number of monthly searches, and optionally an Apify token for more accurate view counts. You then install the separate skill file through Claude's settings, and finish first time setup by signing into a secondary Google account, since the tool keeps cookies stored locally, followed by creating and sharing a NotebookLM notebook link. Once set up, a user simply types a research request in plain language, describing a topic, a minimum view count, and a time window, and the tool searches for matching videos, adds each one to the shared notebook, has Gemini analyze the real transcripts for trends, keywords, and thumbnail patterns, suggests five video ideas, and writes a full script for whichever idea is chosen. The project is written in TypeScript and released under the MIT license.