Getting started¶
With the server installed and authentication completed, it's time to register your first notebook and try a first query.
Registering a notebook in the library¶
NotebookLM notebooks are not automatically accessible to Claude. You need to register them in the library, a local store that the MCP server consults to know which notebooks are available and when to query them.
To register a notebook you need the notebook's URL on NotebookLM and a description of its content.
How to get the notebook URL:
- Go to notebooklm.google.com
- Open the notebook you want to register
- Click Share in the top right
- Select Anyone with the link
- Copy the resulting URL
The URL has a structure similar to https://notebooklm.google.com/notebook/abc123def456.
Registering the notebook in Claude:
Start a conversation with Claude and describe the notebook naturally:
I have a NotebookLM notebook dedicated to n8n documentation, with tutorials, official guides, and personal notes. I'd like to add it to the library so I can query it. The URL is https://notebooklm.google.com/notebook/abc123def456
Claude asks a few questions to better understand the content and proposes a structured card with name, description, topics, and use cases. After confirmation, the notebook is saved in the local library.
The importance of metadata
The description and topics provided during registration are not mere labels. Claude uses them to decide which notebook to query when it receives a question. An accurate description improves the precision of automatic selection — if you indicate that the notebook contains "tutorials on automation with n8n and API integrations", Claude will know to associate it with questions on those specific topics.
The first question¶
With the notebook registered, you can make your first query. It's not mandatory to specify which notebook to consult since Claude chooses based on metadata, but it's advisable when the registered notebook library contains similar ones.
How do you install n8n locally on Windows 11?
Claude identifies the relevant notebook, creates a research session with NotebookLM, structures the question with source fidelity constraints, and returns the source-grounded response. The entire process happens transparently — the user doesn't need to worry about the structuring, which is applied behind the scenes as described in the How it works chapter.
Reading the structured response¶
The response follows the format imposed by the automatic structuring. Each piece of information provided includes a reference to the document source. If a requested piece of information is not present in the documents, the label [NOT FOUND IN DOCUMENTS] appears.
Trust the [NOT FOUND]
The absence declaration is a positive result, not an error. It means the system is effectively limiting responses to what is documented, without inventing or integrating with external knowledge. This is the expected and desirable behavior, especially when working with technical or legal documentation where precision is essential.
Verifying on NotebookLM¶
NotebookLM saves chats, which provides a direct verification mechanism for what happened during the query:
- Open the notebook at notebooklm.google.com
- Check the chat history
- Read the structured prompt that Claude sent
- Read the original response from NotebookLM with internal reference links to the documents This transparency mechanism allows you to verify that structuring was applied and to compare the original response with the presentation provided by Claude. In contexts where source fidelity is critical, such as legal analysis or technical procedure verification, this audit capability is particularly valuable.
Continuing the conversation¶
After the first question, you can dig deeper without starting over. Claude maintains the session with NotebookLM, which retains the context of previous questions.
What if I wanted to configure an external PostgreSQL database instead of the built-in one?
NotebookLM uses the context of the previous conversation (the Windows 11 installation) to provide a more precise and contextualized response. This progressive approach — starting with a broad question and deepening step by step — is one of the most effective ways to work with the system. Each answer can suggest the next question, building an increasingly detailed understanding of the topic.
Under the hood
When Claude queries NotebookLM for the first time on a notebook, the server creates a research session identified by a unique ID. The session corresponds to a persistent chat in NotebookLM. As long as the session remains active, subsequent questions are sent to the same chat, allowing NotebookLM to use the accumulated context. Sessions have a configurable timeout (15 minutes of inactivity by default) and the maximum number of concurrent sessions is limited to 10.