Companion guide · Static-corpus research
Claude is brilliant at conversation and drafting, but its context window has limits. The Family Law Act runs to hundreds of pages. The Rules, more. A complex matter's disclosure bundle can be thousands. For that volume, you need a different kind of engine — one built to ingest a static corpus once, build an index, and answer questions with citations back to source paragraphs. That's NotebookLM.
Two lenses: what it does, and what you'd put in it.
NotebookLM is a Google product that turns a set of documents (your "sources") into a queryable knowledge base. You upload sources once, then chat with them. Every answer carries citations back to the exact passage in the source it came from. It uses Gemini under the hood.
It is purpose-built for research over a fixed body of material, which makes it complementary to — not a replacement for — your regular AI assistant.
Once it's all loaded, you can ask "where in the rules does it deal with X?", "summarise the other side's position on parenting", "which authorities have addressed gatekeeping?" — and get cited answers in seconds.
Three steps. Free tier is sufficient for most SRL matters.
Sign in with any Google account. Free tier is generous and adequate for most SRL matters. Pro tier (currently included with Google One AI Premium) lifts source-per-notebook limits and adds extras like longer audio overviews.
Don't put everything in one notebook. Recommended split:
Separation keeps queries clean and citations interpretable.
Drag in PDFs, paste in URLs, or import from Google Drive. NotebookLM accepts up to the current per-notebook source limit (varies by tier).
Tip: convert .docx and HTML to clean PDF first — NotebookLM handles PDF best.
After upload, you can:
Where it earns its place, and where it doesn't.
An honest read on whether you need this at all.
NotebookLM probably suits you if…
It probably doesn't if…
Frankly, this tool earns its place in matters with real volume. For a four-document matter, you don't need it. For a four-hundred-document matter, you'll wonder how anyone manages without it.
Use it from inside your assistant, not in a separate tab.
The NotebookLM web interface is good. Using it from inside Claude is better when you're already working on something Claude-related and want to ask the corpus a quick question without context-switching. With an MCP, Claude can query your notebook directly, get the cited answer back, and use it in your current task — drafting, comparing, summarising.
NotebookLM does not yet have a public official API. Existing MCP implementations rely on browser automation against the NotebookLM web app, plus careful handling of Google account state and session cookies. Done well, it's reliable. Done poorly, it breaks every other day.
A maintained instance configured against your Google account, plugged into Claude.
Same monthly as Workspace MCP; one-time setup reflects the extra provisioning.
Setup covers the one-time work of standing up your isolated automation tenant. Once it's running, ongoing cost matches Workspace MCP. Bundle with Workspace MCP: ask about combined pricing on enquiry.
For most SRLs, just use it directly at notebooklm.google.com. The hosted MCP earns its place when you're working with Claude on drafting or analysis and want the assistant to query your corpus mid-task without you having to switch tabs and copy-paste answers back. If you're not using Claude (or another MCP-capable assistant) day-to-day, you don't need the MCP.
Indirectly. NotebookLM can import individual files from Drive at upload time, but it doesn't track live folders. When your sources update, you re-upload. For live Drive access from your assistant, use the Workspace MCP — they're complementary tools.
For finding passages and providing cited summaries of static material it has read, yes. For interpreting that material into legal advice, no — that's not what it is. Use it as a search-and-summary tool over material you've curated. Verify citations and read the underlying passages yourself before relying on anything.
Google's standard terms apply to anything you upload to NotebookLM. The hosted MCP we provide is a broker — it sends your queries to NotebookLM and returns the results to your assistant. We do not store your sources, your queries, or your answers. Request metadata is retained for billing and abuse detection as described in our Privacy Policy.
Both Google (NotebookLM) and OpenAI (with GPTs that can read attached files) offer something in this category. NotebookLM is currently the better fit for the volume and citation-stickiness an SRL needs, and the free tier is more generous. If you're already deep in the OpenAI ecosystem, GPTs with files can substitute. The choice of engine matters less than the discipline of using one consistently.
NotebookLM and the hosted MCP make research over a static corpus faster. They do not make you a lawyer. Substantive legal questions still go to a human admitted to practise — see the landing page FAQ for legal aid pointers if you can't afford private representation.
The pack, install, services, FAQ.
Read / write counterpart for your live Google account.
Where to plug NotebookLM in.