List notebooks visible to the signed-in Google account.

Returns { id, title, icon, source_count, updated_at, url, ownership }. ownership is "mine" (user-created) or "shared" (subscribed/public).

Known limitation: wXbhsf returns the user's most-recent owned notebook only. Bulk pagination for many owned notebooks isn't yet discovered.

Create a brand-new, empty notebook. Returns { id, title, url }.

Uses NotebookLM's CCqFvf RPC, which always creates a fresh notebook (unlike the idempotent "get recent draft" endpoint). title is accepted but NotebookLM's create RPC produces an untitled notebook; it will auto-title once sources are added.

Delete a notebook permanently. Returns { id, deleted: True }.

Irreversible — NotebookLM has no trash/undo. Caller confirms.

Ask a question inside a notebook and return the grounded answer.

The exchange is persisted in the NotebookLM web UI's chat history. Returns { answer, source_count, session_ids, chunks }.

Toggle the public share link on a notebook. Returns the share URL when enabled.

Invite collaborators by email to a notebook.

Add a source to a notebook and return { id, title, word_count }.

• kind="text": value is the raw text to paste. title defaults to "붙여넣은 텍스트" (NotebookLM's own default).

• kind="url": value is a webpage URL. NotebookLM fetches + extracts the content server-side; title is ignored (NotebookLM uses the page's own ).

Not yet implemented: drive, file.

Refresh Drive-linked sources. If source_id is None, refresh all.

Return the extracted text/metadata of a single source.

Start generating a Studio artifact (audio/podcast by default).

Returns immediately with { artifact_id, status, title } — generation is asynchronous and typically takes 2-5 minutes. Poll download_artifact to check progress and retrieve the file when complete.

Currently implemented: audio (Audio Overview / podcast). Other artifact types (video, slides, etc.) raise NotImplementedError.

Revise an existing slide deck (and any revisable artifact) with new instructions.

Check an artifact's status and, if complete, download it to disk.

If artifact_id is None, returns the most recent artifact in the notebook. Returns { artifact_id, status, title, duration_sec, path?, mime? }. path is only present when download succeeded.

Kick off NotebookLM's Discover (Fast Research) flow on a topic.

Returns { query, sources: [{ url, title, description }, ...] }.

The candidate sources are NOT automatically added to the notebook. Use source_add(kind="url", value=<chosen_url>) to import selected ones.

Requires an EMPTY notebook — NotebookLM disables Discover once the notebook has content. Create one with notebook_create first if needed.

One-shot: create ephemeral notebook → research → import sources → Q&A.

Full workflow wrapped in a single call:

1. Create a fresh empty notebook.

2. research_start(topic) → candidate URLs.

3. Import the top max_sources URLs as sources.

4. For each question in questions, run notebook_query.

5. If keep_notebook is False (default), delete the notebook.

Returns { notebook_id, # str if kept, None if deleted topic, sources_imported, # [{url, title, word_count}, ...] qa, # [{question, answer, source_count, chunks}, ...] kept, # bool }

Ask the same question across multiple notebooks and return per-notebook answers.

Returns { question, results: { : { answer, source_count, session_ids, chunks } | { error: }, ... }, ok_count, error_count, }

Per-notebook failures (no sources, RPC errors) are captured in error so one bad notebook doesn't abort the fan-out.

Run a single tool operation across many notebooks (e.g. operation="notebook_query").

Run a sequence of tool invocations. Each step: { tool, params }.

Output of step N is available as $prev inside step N+1's params.

Add/remove tags on one or more notebooks. mode in {"add","remove","set"}.

Re-validate the stored Google session. Returns { logged_in: bool, email?: str }.