NotebookLM + Claude via MCP:
The Ultimate Research Stack

Create Topic-Specific Notebooks

Don't dump everything into one notebook. Create focused notebooks by project or research question:

• AI Ethics Research — Hiring Algorithms — papers, regulatory docs, audit reports
• Competitor Analysis Q1 2026 — earnings transcripts, product docs, press releases
• Dissertation Ch.3 — Methodology — methodology papers, exemplar studies, notes

Rule of thumb: If you'd need a separate literature review section for it, it deserves its own notebook.

Optimize Source Quality

NotebookLM supports up to 50 sources per notebook (300 on Plus). For MCP workflows, quality beats quantity:

• PDFs: Clean, OCR'd documents work best. Scanned images with poor OCR will pollute results
• YouTube: Upload key lectures and talks — NotebookLM extracts and indexes the transcript
• Web URLs: Use for live reference materials, but verify the page isn't paywalled
• Google Docs/Slides: Great for your own notes and working documents

Pre-Process Before Connecting Claude

Before you bring Claude into the loop, use NotebookLM's native features to create structured artifacts:

• Generate a Briefing Doc — gives Claude a high-level map of your sources
• Create Pinned Notes with your research questions and hypotheses
• Run exploratory queries to verify your sources are being parsed correctly
• Use the Mind Map feature to visualize the conceptual landscape

Install the MCP Server

The community-built NotebookLM MCP server enables Claude to query your notebooks directly:

# Clone the MCP server repository
git clone https://github.com/notebooklm-mcp/server.git
cd server

# Install dependencies
npm install

# Configure your Google account credentials
cp .env.example .env
# Edit .env with your Google account details

Important: Use a separate Google account if privacy is a concern. This is a community-built tool — treat it like any developer tool in your stack.

Configure Claude Desktop

Open Claude Desktop's MCP settings and add the NotebookLM server:

{
  "mcpServers": {
    "notebooklm": {
      "command": "node",
      "args": ["path/to/server/index.js"],
      "env": {
        "GOOGLE_AUTH_TOKEN": "your-token-here"
      }
    }
  }
}

Restart Claude Desktop. You should see an icon indicating the MCP connection is active. Claude can now query your NotebookLM library.

Verify the Connection

Test with a simple query to confirm everything works:

Look at my NotebookLM notebooks and list the ones available.
Then summarize the key sources in my [notebook name] notebook.

If Claude returns your notebook names and source summaries, you're connected.

Use NotebookLM Alone When:

• You need a quick summary of a single document
• You want an Audio Overview podcast for commute listening
• You need flashcards, quizzes, or study guides from your material
• You want to verify a specific fact exists in your sources

Use Claude Alone When:

• You need creative writing, brainstorming, or code generation
• You're working with general knowledge (no private sources needed)
• You need multi-step reasoning about abstract concepts
• You want real-time web search via Claude's web tools

Use Both via MCP When:

• You need Claude's reasoning applied to YOUR specific documents
• You're synthesizing across 10+ sources and need original analysis
• You want production-grade written output grounded in real evidence
• You're doing competitive intel, literature reviews, or strategic analysis
• You need to continuously query your research while building an argument