NotebookLM Operations Skill

NotebookLM Operations

Purpose

Route NotebookLM tasks to the correct interface (@notebooklm-mcp, nlm CLI, or notebooklm CLI) with explicit profile safety (personal/work) and predictable operational flow. Two clients are installed, exposing three interfaces: the notebooklm-mcp-cli package provides both the nlm CLI and the @notebooklm-mcp MCP server; the notebooklm-py package provides the notebooklm CLI for pure-Python automation.

Do not enable @notebooklm-mcp unless this skill has been invoked first and the decision guide below points to it.

Trigger Phrases

Invoke this skill when you receive any of the following:

"ingest sources into NotebookLM"
"generate a podcast / audio overview"
"create a study guide / briefing doc / FAQ"
"sync these docs to NotebookLM"
"what notebooks do I have?"
"create/delete/rename a notebook"

Decision Guide

| Use Case | Tool | Why | |----------|------|-----| | Interactive chat, one-off notebook ops, query a notebook | @notebooklm-mcp (toggle on) | 35 MCP tools; real-time | | Scripted / headless notebook ops, batch source upload | nlm CLI (notebooklm-mcp-cli) | Same API, no MCP server overhead | | Pure-Python automation, CI pipelines | notebooklm CLI (notebooklm-py) | Lightweight Python API; no browser/Chrome | | Discover notebooks, check source counts | nlm notebook list --profile <p> | Fastest read-only lookup | | Auth troubleshooting, health check | nlm doctor --verbose | Covers both clients |

After finishing an @notebooklm-mcp session, toggle it back off to avoid bloating future sessions' tool lists.

Profiles

Two accounts are configured. Always pass --profile explicitly; never mix sources across accounts. To inspect the actual email/account mapped to each profile in your local environment, run nlm doctor --verbose or nlm profile list.

| Profile | Account | Use for | |---------|---------|---------| | personal | <personal-email> | Personal research, open-source work | | work | <work-email> | Work/client projects, internal docs |

# List notebooks per profile
nlm notebook list --profile personal
nlm notebook list --profile work

# Inspect actual identity per profile
nlm doctor --verbose

Pipelines

1. ingest → podcast

Add sources then request an audio overview.

# Create notebook (if new)
nlm notebook create --title "My Research" --profile personal

# Add sources (URLs, files, or Drive)
nlm source add <notebook-id> --url https://example.com/paper.pdf --profile personal
nlm source add <notebook-id> --file ./local-doc.md --profile personal

# Generate audio overview (async — poll for status)
nlm studio create <notebook-id> --type audio-overview --profile personal
nlm studio status <notebook-id> --profile personal

Via @notebooklm-mcp (toggle on):

mcp__notebooklm-mcp__notebook_create  → create notebook
mcp__notebooklm-mcp__source_add       → add each source
mcp__notebooklm-mcp__studio_create    → request audio overview
mcp__notebooklm-mcp__studio_status    → poll until ready
mcp__notebooklm-mcp__download_artifact → download the mp3

2. research → report

Add sources then generate a study guide or briefing doc.

nlm notebook create --title "Topic Research" --profile work
nlm source add <notebook-id> --url https://... --profile work
nlm pipeline <notebook-id> study-guide --profile work
# or
nlm pipeline <notebook-id> briefing-doc --profile work

Via @notebooklm-mcp:

mcp__notebooklm-mcp__notebook_create
mcp__notebooklm-mcp__source_add        (repeat per source)
mcp__notebooklm-mcp__pipeline          --type study-guide | briefing-doc | faq
mcp__notebooklm-mcp__export_artifact   → save output locally

3. multi-format batch

Mixed URL + local file sources into one notebook.

# Add URLs and files in one batch call
nlm batch add-sources <notebook-id> --profile personal \
  --urls "https://url1.com,https://url2.com" \
  --files "./doc1.md,./doc2.pdf"

Via @notebooklm-mcp:

mcp__notebooklm-mcp__batch   → accepts mixed source list

MCP Hygiene

@notebooklm-mcp exposes 35 tools. When it is enabled those tools appear in every tool invocation in the session, adding latency and context weight.

Protocol:

Toggle @notebooklm-mcp on at the start of a NotebookLM task (/mcp → enable notebooklm-mcp).
Complete all NotebookLM operations.
Toggle @notebooklm-mcp off before switching to unrelated work.

If @notebooklm-mcp is not listed in your Claude Code MCP panel, run:

nlm setup add claude-code
# then restart Claude Code

Troubleshooting

Run nlm doctor --verbose first. It checks installation, auth cookies, CSRF token, Chrome profiles, and AI tool config in one pass.

| Symptom | Code | Recovery | |---------|------|----------| | Auth failure / 401 | -32030 | nlm login --profile <p> or re-run nlm setup add claude-code | | Notebook not found | -32031 | Verify ID with nlm notebook list --profile <p> | | Source limit reached (50 max) | -32032 | Delete old sources before adding new ones | | Unsupported format | -32033 | Convert to PDF/MD/TXT first | | Upload failed | -32034 | Check file size (< 200 MB) and retry | | API error | -32035 | Check nlm doctor --verbose; retry after 60 s |

Error codes -32030–-32039 are reserved for sync-to-notebooklm. Full registry: ../../docs/error-codes-registry.md.

Related Specs

../../docs/specs/sync-skill-interface.md — standard interface all sync-to-* skills must implement
../../docs/error-codes-registry.md — MCP-JSON-RPC error code registry; -32030–-32039 reserved for NotebookLM
Phase 3 (deferred): sync-to-notebooklm skill — automated doc sync using the reserved error range; tracked as Linear VKO-89 / VKO-90 (labels: research, automation)

MAOS Community Skill v1.1.0 | OS-agnostic, project-agnostic

Agent Skills: NotebookLM Operations

Install this agent skill to your local

Skill Files