What NotebookLM is good at
NotebookLM is one of the best tools for asking grounded questions about a fixed set of sources. You upload documents into a notebook, and it answers with citations back to those sources, plus extras like audio overviews. For literature review and source-grounded Q&A, it is genuinely strong.
Where it falls short as a knowledge base
NotebookLM is read-first. You feed it sources and ask questions; it is not a place you accumulate and edit your own durable notes. It runs on Google's Gemini inside its own product, so Claude and ChatGPT cannot use what is in it. And it has no official MCP server or write API. The NotebookLM MCP servers you find online are unofficial community projects that automate the web UI, which can break when the UI changes.
The alternative depends on the job
If what you actually want is a knowledge base you write to, that every assistant can read and update, that is Hjarni: plain Markdown notes with a built-in MCP server, readable by Claude, ChatGPT, Cursor, and more. If your job is really "chat over hundreds of uploaded documents," a managed RAG service like Ragie or Pinecone Assistant is the closer match, and you can even connect both to the same assistant.
NotebookLM answers questions about documents you upload. Hjarni is the memory you write to, that every assistant can read.
When NotebookLM is still the right tool
If you mainly need citation-backed answers over a specific set of PDFs and articles, and audio overviews are a bonus, NotebookLM is hard to beat for that. Many people keep it for source Q&A and use a writable knowledge base for the notes and decisions their assistants should share across every conversation.