Notebooklm

NotebookLM CLI

Wraps Google NotebookLM through the nlm command-line tool (notebooklm-mcp-cli package, installed via uv tool install). Talks to NotebookLM's internal API using cookies from a real browser login - no API key required.

Authentication (check first)

Sessions last ~20 minutes. Before any operation, verify:

nlm login --check

If it reports "not authenticated" or "cookies expired", the user has to run nlm login themselves in their terminal - it opens a browser for Google sign-in and extracts cookies. Tell them to type ! nlm login in the prompt (the ! runs it in the live shell so output lands here). Do not try to launch it from a sub-agent or background bash - it needs an interactive browser.

Profiles for multiple Google accounts: nlm login --profile work, then nlm login switch work to make it default. List with nlm login profile list.

Auth state and cookies live in ~/.notebooklm-mcp-cli/.

Command structure

Two equivalent forms - pick whichever reads cleaner:

nlm notebook create "My Project"        # noun-first
nlm create notebook "My Project"        # verb-first

--json or --quiet for machine-readable output. --confirm skips destructive-action prompts (required for any non-interactive call).

Common workflows

Ingest URLs and generate a deep-dive podcast

# `notebook create` prints "ID: <uuid>" - extract it:
NB=$(nlm notebook create "Topic XYZ" | awk '/^  ID:/ {print $2}')
nlm source add $NB --url "https://example.com/article-1" --wait
nlm source add $NB --url "https://example.com/article-2" --wait
nlm audio create $NB --format deep_dive --length long --confirm
# Poll until ready, then download (artifact ID goes via --id, NOT positional):
nlm studio status $NB
nlm download audio $NB --id <artifact-id> --output podcast.mp3

nlm download <type> <notebook> alone grabs the latest artifact of that type. Pass --id to pick a specific one.

--wait matters - sources must finish processing before any artifact generation will use them. Note: --json works on list-type commands (nlm notebook list --json) but not on create; parse stdout for the ID, or alias the notebook with nlm alias set my-nb <id> once you have it.

Briefing doc / study guide / blog post from existing notebook

nlm report create $NB --format "Briefing Doc" --confirm
nlm report create $NB --format "Study Guide" --confirm
nlm download report $NB --id <artifact-id> --output briefing.md

Report formats: "Briefing Doc", "Study Guide", "Blog Post", "Create Your Own" (pass with --description "...").

One-shot pipeline (ingest + multiple artifacts)

nlm pipeline list                                                # see built-ins
nlm pipeline run $NB ingest-and-podcast --url "https://..."
nlm pipeline run $NB research-and-report --url "https://..."
nlm pipeline run $NB multi-format

Custom pipelines: drop YAML in ~/.notebooklm-mcp-cli/pipelines/.

Query / chat across notebooks

nlm notebook query $NB "What are the three biggest risks?"
nlm cross query "Compare approaches" --notebooks "id1,id2"
nlm cross query "Summarize all" --tags "ai,research"

Aliases for IDs you reuse

nlm alias set my-research <notebook-id>
nlm notebook query my-research "..."

Studio artifact types (cheat sheet)

All take --confirm for unattended use. Audio/video typically takes 1-5 minutes - poll with nlm studio status $NB (or --wait).

Source types

nlm source add $NB --url "https://..."           # web page
nlm source add $NB --youtube "https://..."       # YouTube transcript
nlm source add $NB --file path/to/document.pdf   # local file (PDF, etc.)
nlm source add $NB --text "raw content" --title "Notes"
nlm source add $NB --drive <google-doc-id>       # requires Drive access on the same Google account
nlm source sync $NB --confirm                    # re-fetch Drive sources
nlm source stale $NB                             # show sources needing sync

Add --wait whenever the next step depends on the source being processed.

Saving downloads

Download artifacts into the output/ directory under the current working directory:

mkdir -p output
nlm download audio $NB --id <art-id> --output output/podcast.mp3
nlm download report $NB --id <art-id> --output output/briefing.md

Gotchas

• Free tier ~50 queries/day. Batch and pipeline commands burn through this fast.
• Always pass --confirm for create/delete from a script - the CLI prompts otherwise and will hang.
• Always pass --wait to source add when the next step uses that source.
• Unofficial API - the package wraps NotebookLM's internal endpoints, so behavior can change. If something breaks unexpectedly, run nlm doctor first.
• Rate-limit safety: prefer nlm cross query --notebooks "id1,id2" over running the same query N times.
• German output: NotebookLM follows source language. If sources are German, generated artifacts will be German - real umlauts come through fine, no transliteration concerns.

Diagnostics

nlm doctor                  # checks install, auth, browser, MCP setup
nlm doctor --verbose

MCP server (optional)

The same package ships an MCP server (notebooklm-mcp). To expose NotebookLM as MCP tools to Claude Code at user scope:

nlm setup add claude-code

This skill assumes CLI usage by default. If MCP tools named mcp__notebooklm-mcp__* appear in the tool list, prefer those (they avoid shell quoting).

Full reference

references/cli_reference.md is the verbatim nlm --ai dump (every command, every flag, every recovery flow). Read it when a non-obvious command or edge case comes up. Or run nlm --ai directly to see the latest version after upgrades.

Install / upgrade:

uv tool install notebooklm-mcp-cli       # first install
uv tool upgrade notebooklm-mcp-cli       # update