NotebookLM MCP Server by PleasePrompto

Register a NotebookLM notebook in the local library so it can be queried with ask_question, ingested into with add_source, etc.

Required URL

The user must supply a NotebookLM share-link. To produce one:

1. Open https://notebooklm.google
2. Open the notebook → click "Share" (top right)
3. Set "Anyone with the link" → "Copy link"

Permission workflow

Do NOT call this tool unprompted. The expected dialogue is:

1. Ask for the URL
2. Ask what knowledge it contains (1–2 sentences) → description
3. Ask which topics it covers (3–5) → topics
4. Ask when it should be consulted → use_cases
5. Propose a name and the metadata back to the user
6. Only after explicit confirmation, call add_notebook.

Free-tier limits: 100 notebooks · 50 sources each · 50 queries/day. Google AI Pro/Ultra raises these 5×.

Ingest a source into a NotebookLM notebook. Supports two source types in v2.0:

• url — NotebookLM crawls and indexes a website
• text — paste raw text (treated as a copied document)

File / YouTube / Google-Drive uploads are not yet implemented.

Returns sourceCountBefore/sourceCountAfter so the caller can verify the new source landed. Call once per source — multiple sources require multiple calls. NotebookLM finishes indexing within 5–30 seconds; subsequent ask_question calls then have the new source in context. Free notebooks cap at 50 sources.

Known quirk: pasted-text uploads occasionally redirect to a freshly created "Untitled notebook" on Google's side. The tool detects this and returns a clear error so you can re-try against the correct URL.

Conversational Research Partner (NotebookLM • Gemini 2.5 • Session RAG)

No Active Notebook

• Visit https://notebooklm.google to create a notebook and get a share link
• Use add_notebook to add it to your library (explains how to get the link)
• Use list_notebooks to show available sources
• Use select_notebook to set one active

> Auth tip: If login is required, use the prompt 'notebooklm.auth-setup' and then verify with the 'get_health' tool. If authentication later fails (e.g., expired cookies), use the prompt 'notebooklm.auth-repair'.

Tip: Tell the user you can manage NotebookLM library and ask which notebook to use for the current task.

Two-phase deep cleanup of all server data on disk (auth state, browser profiles, caches, MCP logs, temp backups). Cross-platform (Linux/macOS/Windows). Always close all Chrome/Chromium instances first — open browsers can lock files.

Phase 1 (preview): call with confirm: false. Returns a categorised list of paths and total size. No deletion happens. Phase 2 (delete): after the user reviews the preview and approves, call with confirm: true.

Set preserve_library: true to keep the notebook library file (library.json) while wiping everything else — recommended when troubleshooting auth.

Permanently close a session and discard its browser tab. Use the id returned by list_sessions (or the session_id from a prior ask_question response). Closed sessions cannot be resumed — start a new one with ask_question (no session_id) or pick another from list_sessions. Ask the user before closing if the session might still be needed.

Save the completed Audio Overview to disk as a .m4a file. Pre-condition: get_audio_status must report status: "ready". Calling this before generation completes returns an error message explaining what to do.

The file lands in destination_dir with NotebookLM's suggested filename (sanitised — usually the audio's title with underscores). The full saved path is returned in result.filePath.

Trigger podcast-style Audio Overview generation for a notebook.

Async by default — returns immediately with one of:

• status: "started" — generation just kicked off
• status: "in_progress" — a generation was already running; this call attached to it
• status: "ready" (with alreadyExisted: true) — an Audio Overview already existed; nothing was triggered

Generation typically takes 2–10 minutes. Workflow:

1. generate_audio → returns immediately
2. Poll get_audio_status every ~30 s
3. When status is ready, call download_audio

Pass wait_for_completion: true for legacy synchronous behaviour (blocks for up to timeout_ms). Audio Overview is the only Studio output exposed in v2.0 (Video / Mindmap / Quiz / Infographic / Datatable / Presentation are NotebookLM features but not yet wrapped).

Non-blocking probe for the current Audio Overview state of a notebook.

Returned status values:

• ready — Audio Overview is generated and ready to download
• in_progress — generation is currently running
• not_started — no Audio Overview exists yet for this notebook

Safe to poll every ~30 s while waiting for generate_audio to finish. When status flips to ready, call download_audio with a destination directory.

Inspect server state. Returns:

• authenticated — whether saved Google cookies are still valid
• notebook_url, active_notebook_id, active_notebook_name — the currently selected library notebook (or null)
• total_notebooks — library size
• active_sessions, max_sessions, session_timeout — runtime session stats (timeout in seconds; sessions auto-close after this)
• headless, auto_login_enabled, stealth_enabled — config

Use this first thing in a new conversation. If authenticated=false, run setup_auth (or re_auth to switch accounts).

Aggregate statistics about the local notebook library: total_notebooks, active_notebook (id), most_used_notebook, total_queries, last_modified. Useful as a quick health check or when the user asks "what notebooks do I have?".

Fetch full metadata for one notebook by id. Use to verify what's currently stored before calling update_notebook, or to show the user the exact description/topics/use_cases Claude has for it.

List every notebook in the local library with its metadata (id, name, description, topics, tags, url, use_count, etc.). Use the returned id for select_notebook, update_notebook, get_notebook, remove_notebook, or as notebook_id on ask_question / add_source / audio tools.

List all active browser sessions for this server. Each entry includes id, created_at, last_activity, age_seconds, inactive_seconds, message_count, and notebook_url. Useful before reset_session / close_session or to recover a session_id you can pass back into ask_question to continue an existing conversation. Sessions older than session_timeout (see get_health) are auto-closed.

Switch to a different Google account or recover from broken auth. Closes all active sessions, deletes saved cookies and Chrome profile, and opens a fresh login browser.

Common triggers:

• NotebookLM's 50 queries/day free-tier limit is reached and the user wants to rotate to another Google account
• setup_auth failed and a clean slate is needed

After login, call get_health to verify. For very stuck states, run cleanup_data(confirm=true, preserve_library=true) before re_auth.

Remove a notebook from the local library. Does NOT delete the actual NotebookLM notebook on Google's side — only the local metadata entry. Active sessions on this notebook are closed.

Confirmation workflow: look up the notebook by id, ask the user "Remove '[name]' from your library?" — only call after a clear yes.

Clear a session's chat history while keeping the same session id. Use this when the task changes mid-conversation and you want a fresh context without losing the underlying browser tab. Ask the user before resetting if they might still need the prior history.

Search the library by free-text query — matches against name, description, topics, and tags. Returns notebook objects with their id so you can chain into select_notebook etc.

Use this when the user references a notebook by topic ("the React one") instead of by exact name. If multiple notebooks match, propose the top 1–2 and let the user choose.

Mark a notebook as the active default. After this, ask_question, add_source, and the audio tools resolve to that notebook when the caller omits notebook_id / notebook_url.

When to call:

• The user explicitly switches context (e.g. "Let's work on React now")
• Task obviously needs a different notebook than the current one — announce the switch ("Switching to the React notebook…") before calling.
• If the right notebook is ambiguous, ask the user first instead of guessing.

Open a browser window for first-time Google login. Returns immediately after spawning the browser; the user has up to 10 minutes to complete sign-in, then cookies are persisted for future runs.

When to use:

• get_health reports authenticated=false for the first time
• Auto-login credentials are not configured
• re_auth is the right call when you want to *switch* accounts or recover from a daily-quota lockout

After login finishes, call get_health to verify success.

If the browser session feels broken (auth keeps failing, stale cookies), run cleanup_data(confirm=true, preserve_library=true) first, then retry setup_auth.

Patch metadata fields on an existing notebook. Pass id plus any subset of name, description, topics, content_types, use_cases, tags, url — only supplied fields change.

Workflow: identify the target notebook → propose the exact change back to the user → call only after explicit confirmation. Multiple fields can be updated in one call.