Paolo Dalprato
Connecting Claude and NotebookLM¶
Last updated: March 15, 2026
Claude is a versatile tool for analyzing texts, answering questions, and generating content. However, when the work requires managing extensive documentation — dozens of documents, or just a few but very long ones — the practical limits of the context window become apparent. Google's NotebookLM addresses this problem in a complementary way, allowing you to upload large amounts of material and query it with source-grounded answers.
This manual documents NotebookLM MCP Structured, an MCP server I developed to connect Claude Desktop to NotebookLM. The integration allows you to query document collections loaded in NotebookLM directly from a conversation with Claude, combining Claude's expressive capabilities with NotebookLM's document fidelity.
Who this guide is for¶
The guide is designed for Claude Desktop users who need to work with extensive documentation: technical manuals, regulatory collections, research corpora, corporate archives. No programming skills are required, but you need to be able to run a few commands in the terminal by following step-by-step instructions.
What you need¶
To use the system you need three components: a Claude Pro subscription (or higher) with Claude Desktop installed, a Google account (free or paid) for NotebookLM, and Node.js version 18 or later installed on your computer. The Prerequisites chapter describes in detail how to verify and obtain each component.
What the manual covers¶
The first chapters explain why the integration is useful and how it works, with particular attention to the problem of source fidelity and the automatic structuring solution. The middle chapters walk you through installation, authentication, and initial operations. The final chapters cover notebook library management, querying strategies, real-world use cases (including a comparative legal analysis that demonstrates the method in action), and troubleshooting.
The project¶
NotebookLM MCP Structured is a fork of the notebooklm-mcp MCP server by Gérôme Dexheimer, who made the connection between Claude and NotebookLM possible. My fork adds an automatic prompt structuring system that transforms user questions into structured requests with explicit source fidelity constraints before sending them to NotebookLM. On the return path, the system instructs Claude to present the response without adding its own knowledge that doesn't come from the documents.
The project is open source, available on GitHub.
Under the hood
The fork was developed entirely through vibe coding with Claude: design and code writing with Sonnet 4.5, review and refactoring with Opus 4.6. No line of code was written manually — a concrete example of how generative AI tools can be used to develop working software starting from problem definition and dialogue with the model.