What Is notebooklm-coach? Claude Code Skill Setup, Is It Free, Is It Safe
You load a YouTube channel's videos into a NotebookLM notebook as sources from the terminal; then you ask questions from Claude Code, and each answer links back to the passage in the video. It is not a Google or Anthropic product: it is a community repository published by a developer named Artem Novitckii as a fixed fork of another skill (4 commits, 70 stars, last updated August 17, 2026). It is MIT-licensed and free; NotebookLM needs a Google account, and a free account has a limit of 50 sources per notebook. Watch out: it connects to NotebookLM through two unofficial command-line tools, using the Google session cookies in your browser.
- What it does
- Loads a YouTube channel into NotebookLM so you can query it from Claude Code
- Type
- Skill (community, unofficial)
- Google access
- Unofficial client, via cookies
- Where
- Claude Code (macOS/Linux)
- Free limit
- 50 sources / notebook
- Turkish
- Not stated in the documentation
- Cost
- Free, MIT
What is notebooklm-coach for?
NotebookLM (now called "Gemini Notebook" on Google's help pages) is a notebook tool that answers based on your own sources. This skill is used to bulk-load an expert's YouTube channel into such a notebook and ask your questions from Claude Code. The workflow the README describes: load up to the 300 most recent videos from the channel, ask a question, and get citations that link the answer to the passage in the video.
!This is not an official Google or Anthropic tool. It is a single developer's fork: 4 commits, no tagged release, no issues or pull requests. The README says the NotebookLM interface does not let you add a channel in bulk; the skill fills that gap from the terminal.
Who it helps: people comfortable in the terminal who want to pull information out of an expert's video archive through questions and answers, without watching it end to end. If you want, answers are written to an Obsidian vault with source links (optional).
How do you install notebooklm-coach in Claude Code?
Requirements: Claude Code, a Google account and macOS or Linux. On any other operating system the script (bootstrap.sh) stops with the message "run inside WSL"; if you use Windows, this route needs WSL. The README and SKILL.md don't mention Windows.
- 1. Download into the skills folder: Run the git clone command from the README (below).
- 2. Install the requirements: bootstrap.sh installs uv, two command-line tools (nlm and notebooklm) and the Playwright Chromium browser (about 100 MB).
- 3. Do the two logins yourself: nlm login and notebooklm login open a browser for Google sign-in; the README says not to have Claude Code run them, but to run them yourself.
- 4. Verify: If the nlm notebook list command returns an empty list ([]), the login works.
› git clone https://github.com/artemnovitckii/notebooklm-coach.git ~/.claude/skills/notebooklm› bash ~/.claude/skills/notebooklm/scripts/bootstrap.sh› nlm login› notebooklm loginAfter installing, restart Claude Code; the README says the skill activates on phrases such as "notebooklm" and "load channel". If uv is missing, bootstrap.sh runs Homebrew or astral.sh's installer (a script downloaded with curl) and installs the two tools from PyPI without pinning versions: if you want to read the script before running it, the file is scripts/bootstrap.sh in the repository.
This guide is based on the official README, SKILL.md and the scripts.
How do you load a YouTube channel and ask questions with notebooklm-coach?
Write your request to Claude Code as usual; the skill activates on phrases such as "load channel". The workflow below comes from the README's quick example; the channel URL, notebook ID and count will differ for you.
› python3 ~/.claude/skills/notebooklm/scripts/load_channel.py scrape --channel "https://www.youtube.com/@hubermanlab" --output /tmp/huberman-videos.json› notebooklm create "Andrew Huberman - Health"› uv run --with "notebooklm-py[browser]" python3 ~/.claude/skills/notebooklm/scripts/load_channel.py load --videos /tmp/huberman-videos.json --notebook <notebook-id> --count 300 --concurrency 6› nlm notebook query <notebook-id> "What does Huberman recommend for sustaining deep focus?" --jsonThe repository says 300 videos load in 3 to 5 minutes at concurrency 6 (the maker's own measurement). After loading, NotebookLM processes the videos on its side; the README says the first queries may be weak for a few minutes. If you later want to write the same work to Obsidian, the scripts in SKILL.md (resolve_citations.py and others) write the answers as notes with links to the source passage; they are run from the vault folder and require the Dataview plugin.
Where does notebooklm-coach run?
| Claude CodeYes | The README and SKILL.md only describe this route. |
|---|---|
| macOS, LinuxYes | bootstrap.sh accepts these two systems. |
| WindowsNeeds WSL | The script says "run inside WSL"; whether it works outside WSL is not stated. |
| Claude.ai, Cursor, CodexUnclear | Not stated in the documentation; that doesn't mean it won't work. |
| Turkish videos and Turkish questionsUnclear | Not stated in the documentation; the answer language depends on NotebookLM. |
What does notebooklm-coach access? Is it risky for my Google account?
!The skill connects to NotebookLM not through an official Google route but through two third-party tools. One borrows the cookies of the Google session open in your browser; the other opens its own Chromium browser and asks for a Google sign-in. The README of notebooklm-py, which runs in the background, says it uses undocumented Google APIs, is not affiliated with Google and may break at any time. No source says whether Google will restrict your account for this kind of access. Consider a separate account instead of trying it with an important one.
Notebooks, the source list and source contents; the YouTube channel page (no login).
Creates notebooks in your Google account and adds video URLs as sources. Locally it writes a browser profile and session data under ~/.notebooklm and temporary files under /tmp, and the Obsidian scripts write Markdown to your vault.
The skill itself does not delete anything. The README gives an example delete command for you to run to clean up red sources.
The skill files make no calls to a server of their own. Network: Google (via cookies) and YouTube; during setup, PyPI and, if needed, astral.sh.
Security check (October 15, 2026): the GitHub Security Advisories list for the skill repository is empty; the dependency notebooklm-mcp-cli has 4 entries (2 medium, 2 low), all fixed in 0.11.2 or earlier; the current version on PyPI is 0.15.5, and bootstrap.sh does not pin versions. OSV has no entries for the two PyPI packages; since the skill itself is not a package, no OSV query applied to it. This does not mean "safe".
Is notebooklm-coach free? What is the source limit?
The skill is free: MIT-licensed (the original skill is MIT too), with no plans or fees in the repository. But because the work runs through NotebookLM, your Google account and its plan limits apply.
50 sources per notebook. If you load more, according to the README the rest become red, empty rows.
On Google's plans page, the Sources row reads 50, 100, 300, 500 and 600 (per notebook). For details by plan, see the Gemini Notebook guide. The README's "Plus and Pro 300" statement does not match that page.
The LICENSE file is MIT. GitHub shows "Other" because it can't detect the license automatically; that does not mean "no license".
Common notebooklm-coach problems
| What happens | What to do |
|---|---|
| I loaded hundreds of videos but only ~50 are filled; the rest are red and empty | The free account limit is 50. On the free plan, load 50 videos or split a large channel across several notebooks. |
| The row title is a raw URL, not the video name (red) | The additions failed; the README says the number grows as concurrency goes up. Delete the red rows and add them again with --concurrency 1. |
| nlm commands return an authentication error | The nlm session is short (README: about 20 minutes). Run the nlm login command again. |
| Reading and writing ended up on different accounts | Check which account each one is on with nlm login profile list and notebooklm auth check; the README says the two logins can land on separate accounts. |
How do you uninstall notebooklm-coach?
The README has no official uninstall step.
- Skill: Delete the Claude Code skill folder (~/.claude/skills/notebooklm).
- Session: Close the write-side session with the notebooklm auth logout command mentioned in the README.
- Google account: Check your Google account security settings for sessions left open on your account; the repository says nothing about this.
Frequently asked questions
What is notebooklm-coach?
A community-made skill that loads a YouTube channel's videos into NotebookLM from the terminal and lets you ask cited questions from Claude Code. It is not an official Google or Anthropic product.
Is notebooklm-coach free?
The skill is MIT-licensed and free. NotebookLM needs a Google account; on the free plan there is a limit of 50 sources per notebook.
Is it risky for my Google account?
The README says it connects to NotebookLM with two unofficial clients; one uses the Google cookies in your browser. Whether an account restriction may happen is not stated in the sources; consider a separate account instead of trying it with your important one.
Does notebooklm-coach work on Windows?
The README and SKILL.md don't mention Windows. Outside macOS and Linux, the setup script stops with the message "run inside WSL".
Is this an MCP?
No; the skill consists of Markdown instructions and scripts for Claude Code and runs the nlm and notebooklm command-line tools. The nlm tool's package name contains "mcp-cli"; that package's own README also describes installing it as an MCP, but this skill uses it only as a command-line tool.
Related guides
Sources
- notebooklm-coach (GitHub, README)
- SKILL.md
- LICENSE (MIT)
- scripts/bootstrap.sh
- notebooklm-py (unofficial client warning)
- notebooklm-mcp-cli security advisories
- Gemini Notebook plans (Google Help)
Not clear in the official sources: Whether Google restricts accounts for access through unofficial clients · running on Windows outside WSL · agents other than Claude Code (Claude.ai, Cursor, Codex) · results with Turkish videos and Turkish questions · which videos can't be added (no captions, region-restricted) · the README's "about 96% citation resolution" and "300 videos in 3-5 minutes" claims (the maker's own measurement) · an official uninstall step. Each is noted in the relevant section.




