The AI Surfer logo
All Resources

Unlocking Research Agents

BY: THE AI SURFERStart riding the AI wave alongside hundreds of other like minded AI Surfers as we learn together how to live better lives and build better businesses leveraging AI.

Join US Now!

ā±ļø 2 minutes

What if you could spin up a research agent that costs you nothing, runs on your computer, and makes slide decks, audio podcasts, quizzes, and even videos in minutes?

That's what happens when you mix Claude Code with NotebookLM.

And it's super easy to set up.

šŸŽÆ What You'll Build:

A research robot. You tell Claude what you need ("read these PDFs and make a slide deck"). Claude does everything through NotebookLM by adding your files, making your stuff, and downloading the results. No clicking buttons yourself. No paying for APIs. Just results.

Primary Resource: šŸ‘‰ notebooklm-py on GitHub

This is an unofficial tool that lets you control NotebookLM from the command line. It can do everything the NotebookLM website can do, plus some things the website can't do yet. It's on PyPI, gets updates often, has an MIT license, and comes with a Claude Code skill file already built in.

What is NotebookLM? NotebookLM (by Google) is a free AI research helper. You give it documents, websites, or text. It reads them and makes summaries, answers questions, and creates polished outputs. The problem: Google doesn't give you an API to control it with code. That's where this automation tool comes in.

This is about finding your North Star. Figure out what you love (the thing only you can do, the thing that lights you up). That's your 10%. Then reverse-engineer AI agents and automations to handle everything else. Inside The AI Surfer, we help you find that North Star, automate the 90%, and spend 100% of your energy on what actually matters.

Prerequisites (2 minutes to check)

  • Google account (to use NotebookLM)
  • Anthropic account (to use Claude Code)
  • Python 3.8 or newer on your computer
  • Terminal access (Mac, Linux, or WSL on Windows)

Claude Code is a command line tool, not a desktop app. You run it from your terminal after installing it. Check Anthropic's docs for the latest install steps.

Pick Your Level

Most people never start. That's what stops them from learning: not the complexity, just the fear of taking the first step. That's why we've broken this down into three levels. Start where you are. Baby steps get you there.

šŸŸ¢ Beginner: Get It Working (15 minutes)

Goal: Install the tool. Run one command. Watch NotebookLM do something automatically.

What to do today:

  1. Install the tool:
bash
pip install notebooklm-py
pip install "notebooklm-py[browser]"
playwright install chromium

The [browser] part installs Playwright. That's the thing that controls NotebookLM's website for you. The chromium command downloads the browser it needs.

  1. Log in to NotebookLM:
bash
notebooklm login

A browser window opens. Log in with your Google account. The tool saves your login on your computer.

  1. Test it:
bash
notebooklm create "Test Research"
notebooklm source add "https://en.wikipedia.org/wiki/Artificial_intelligence"
notebooklm ask "What is AI in one sentence?"

If you see an answer in your terminal, it's working. NotebookLM just read a source and answered you, all from the command line.

Why this matters: You just automated something that would take clicking through NotebookLM's website. Do this with bigger tasks and you're automating entire research jobs.

Next step: Try making something:

bash
notebooklm generate slide-deck

NotebookLM makes a slide deck from your source. Download it and open the PDF. You just went from a URL to a slide deck with two commands.

šŸŸ” Intermediate: Connect It To Claude Code (30 minutes)

Goal: Teach Claude Code how to use NotebookLM. Then you can describe research tasks in normal English and Claude does them.

What to do today:

  1. Get the repo to find the skill file:
bash
git clone https://github.com/teng-lin/notebooklm-py.git
cd notebooklm-py

The repo has a SKILL.md file in the main folder. This teaches Claude Code how to use the notebooklm command.

  1. Set up the skill for Claude Code:

Claude Code finds skills in SKILL.md files in these places:

  • Project-level: .claude/skills/<skill-name>/SKILL.md (in your project folder)
  • Global: ~/.claude/skills/<skill-name>/SKILL.md (works everywhere)

Copy the skill file to your global skills folder:

bash
mkdir -p ~/.claude/skills/notebooklm
cp SKILL.md ~/.claude/skills/notebooklm/

Now Claude Code knows the notebooklm command exists and how to use it.

  1. Test the connection:

Open Claude Code in your terminal and try:

Prompt
Create a NotebookLM notebook called "AI Research", add this Wikipedia article about machine learning as a source: https://en.wikipedia.org/wiki/Machine_learning, then generate a quiz with hard difficulty.

Claude will:

  • Run notebooklm create "AI Research"
  • Run notebooklm source add "https://en.wikipedia.org/wiki/Machine_learning"
  • Run notebooklm generate quiz --difficulty hard
  • Tell you when it's done

You just described a job in normal English. Claude did the commands.

The unlock: You're not typing bash commands anymore. You're describing what you want, and Claude translates it into the right commands. This is where automation starts feeling like having an assistant.

šŸ”“ Expert: Build Multi-Step Research Workflows (1 hour to master)

Goal: Chain together big research tasks (adding multiple sources, asking for insights, making multiple outputs) all done by Claude Code.

What to do today:

  1. Learn what the tool can make:

The tool does way more than slide decks. Here's what you can generate:

bash
notebooklm generate audio "make it conversational" --wait
notebooklm generate video-cinematic
notebooklm generate infographic
notebooklm generate quiz --difficulty hard
notebooklm generate flashcards
notebooklm generate report
notebooklm generate data-table
notebooklm generate mind-map
notebooklm download audio ./research-podcast.mp3

Some of these (like videos and mind maps) aren't even in NotebookLM's website yet. You're using secret features through the hidden API.

  1. Build a real workflow:

Tell Claude Code to do this:

Prompt
I have three PDF files in ~/Documents/research/. Create a NotebookLM notebook called "Market Analysis". Add all three PDFs as sources. Then:
1. Ask: "What are the three biggest trends mentioned across all sources?"
2. Generate a slide deck summarizing those trends
3. Generate an audio overview in an engaging tone
4. Download the audio as podcast.mp3 in the same directory

Wait for each step to complete before moving to the next.

Claude Code will:

  • Go through the PDFs and add them as sources
  • Run the question and get the answer
  • Make the slide deck
  • Make the audio (with --wait so it finishes before moving on)
  • Download the audio file where you asked

You get a slide deck, a podcast, and a text summary from one prompt.

  1. Understand the limits:

Important: This tool uses hidden Google APIs. NotebookLM doesn't officially let you control it with code. That means: APIs could change any time without warning. Rate limits might stop you if you send too many requests. Google could block automation if they notice it. Use this for your personal research work, not for production systems. Check the repo regularly for updates.

Advanced patterns to try:

  • Comparing things: Make separate notebooks for competing products. Ask the same questions in each. Compare what they say.
  • Making things better: Make a draft slide deck. Ask Claude to give feedback. Remake it with the feedback.
  • Multiple formats: From one set of sources, make a slide deck (for presentations), flashcards (for studying), and a report (for documentation).

The shift: At this level, you're not just automating tasks. You're designing research systems. Claude Code becomes the boss. NotebookLM becomes the worker. And you decide what outputs you need.

Why This Combo Is Really Powerful

NotebookLM's strength: It acts like a RAG system (that means it reads your documents and answers questions about them) without you building anything. It makes polished outputs like slide decks, podcasts, and quizzes. And it's completely free.

NotebookLM's weakness: No API. You have to click through the website to use it. That doesn't scale when you're doing big research jobs.

The notebooklm-py tool's strength: It automates all the clicking. You control NotebookLM from the command line. It unlocks features the website doesn't even show yet.

Claude Code's strength: It turns English instructions into terminal commands. You describe what you want. Claude figures out the right commands and runs them.

The combination: You get a research agent that:

  • Costs nothing (NotebookLM is free, Claude Code is pay-per-use but cheap for this)
  • Runs locally (your machine controls everything)
  • Handles the full workflow (from ingesting sources to downloading outputs)
  • Requires zero coding (just describe what you want in plain English)

This is the kind of setup that would have cost thousands to build two years ago. Now it takes two minutes.

You just let AI handle the research grind. Imagine if everything in your life worked like that. What would you do with all that time back? Inside The AI Surfer, we help you automate the 90% so you can live the 100%. Your whole life spent on what lights you up.

Common Issues & Fixes

"notebooklm: command not found"

  • You didn't install it, or Python's bin directory isn't in your PATH. Run pip install notebooklm-py again and check pip show notebooklm-py to see where it installed.

"Authentication failed"

  • Run notebooklm login again. Make sure you're using a Google account that has access to NotebookLM.

"Claude Code doesn't recognize the notebooklm command"

  • The skill file isn't in the right place. Make sure you copied SKILL.md to ~/.claude/skills/notebooklm/SKILL.md. Restart Claude Code after adding it.

"Generation failed" or timeouts

  • NotebookLM can be slow with big sources or complicated generations. Use the --wait flag to make the command wait until it's done instead of timing out.

Next Steps

  1. Start small: Get one source into NotebookLM and generate one deliverable (like a quiz or slide deck) using the CLI.
  2. Add Claude Code: Set up the skill file and try describing a multi-step workflow in plain English.
  3. Go bigger: Chain together multiple sources, multiple questions, and multiple outputs in one prompt.

Check the GitHub repo for the latest features and updates. The tool gets new stuff often.

Ready to automate everything else? Join The AI Surfer where we help you find your North Star, build AI agents to handle the 90%, and spend 100% of your energy on what actually matters. No fluff. Just real automations you can use today.

Last updated: 2026

The AI Surfer logo

Don't Let The AI Wave Crush You!

Start riding the AI wave alongside hundreds of other like minded AI Surfers as we learn together how to live better lives and build better businesses leveraging AI.

Join US Now!