Gemini Notebook (formerly NotebookLM) to Notion & Obsidian

“I manually export or copy the content and reorganize it into my local Obsidian vault.” That’s one answer from the Kortex MCP waitlist survey, and it’s not unusual. In fact, 72 of 776 respondents asked for the same thing. They want notes, answers and Studio outputs to move from Gemini Notebook (formerly NotebookLM) to Notion, Obsidian and Google Docs, where they actually work. Here’s how to hand that job to your AI assistant, so it happens every week without the copy and paste.

Key Takeaways

• 72 of 776 Kortex MCP waitlist respondents asked to get notes and outputs back out of Gemini Notebook (Kortex MCP waitlist survey, 2026).
• Kortex MCP reads your notebooks; your assistant’s own Notion connector, or its file access, does the writing.
• Obsidian is the easiest target. A vault is a folder of Markdown files, and 58% of respondents’ assistants already reach local files, against 30% for Notion.
• Kortex MCP is a Kortex Pro feature in private beta, by invite.

Why can’t NotebookLM connect to Notion or Obsidian directly?

Because Google gives other apps no way in. There’s no public consumer API for Gemini Notebook, and Notion isn’t on its list of source types either (Gemini Notebook Help, 2026). Its own export sends one note at a time to Google Docs, where “any modifications made within the newly created Docs or Sheets file will not synchronize” back (Gemini Notebook Help, 2026). As a result, the usual route looks like this. Open the notebook. Export a note to Docs. Download it as Markdown. Drag it into your vault or import it into Notion. Repeat for the next note, and again next week when the notebook has grown. In other words, every new note costs you the same five steps.

That loop is the norm. Specifically, when we analyzed the 776 survey answers, 244 people (31%) described a manual copy-and-paste workaround they use today (Kortex MCP waitlist survey, Sep-Oct 2026). For this job, we found Obsidian named 16 times and Notion 15 in the free-text answers, more than any other app outside Google’s own. For example, one person downloads each asset to their Downloads folder just so ChatGPT can read it. Another asked to “access assets that I create (mindmaps, videos, etc.)”.

For a one-off move, however, export buttons are enough, and Kortex’s free one-click exports save a notebook’s whole Studio panel as Markdown or a ZIP. For notes you want out every week, in a shape that fits where they’re going, the rest of this guide hands the job to your assistant.

How does your AI assistant become the bridge out?

It reads your notebooks through Kortex MCP and writes to the destination with a connection it already has. Kortex MCP is an MCP server for Gemini Notebook. Your AI client calls it, and Kortex relays the call to the Kortex extension in your own Chrome. The extension then reads your notebooks with the Google session you’re already signed into (Kortex MCP docs, 2026). The writing side, in contrast, uses tools your assistant may already have. Notion’s hosted MCP server lets “AI apps, like Claude, ChatGPT, and Cursor” connect to a workspace (Notion Help, 2026). Coding assistants such as Claude Code, meanwhile, write files straight to disk. Importantly, Kortex never touches Notion or your vault; your assistant does that part, with its own tools.

That split is what makes the result better than an export. The assistant sits between the two apps, so it can reshape content on the way. For example, it can turn a study guide into one Notion page per section, add frontmatter for Obsidian, or drop the notes you don’t need. An export button, by comparison, can’t make those calls.

These are the Kortex MCP tools that do the reading:

What you want out	Kortex tool	What comes back
Find the right notes	list_notes, search_everything	Note titles, types and ids, or matches from across your whole library
Every note in a notebook	get_formatted_content	All notes and artifacts as Markdown or plain text, with no download and no export credits spent
A Studio output	export_artifact	A report as Markdown, a data table as CSV, a media link for audio or video, or a native export to Google Docs or Sheets
Flashcards, quizzes, mind maps	get_artifact_content	The cards, the questions with the correct option, or the mind-map tree
Fresh cited answers	ask_question, batch_to_vault	Answers with citations; batch_to_vault keeps a Markdown file plus a JSON citation sidecar per question
Your Kortex dashboard notes	list_kortex_notes	Notes you wrote in Kortex itself, which Gemini Notebook doesn’t hold

What do you need before you start?

You need an invite first. Kortex MCP is a private beta: only invited members can connect, and invites go out from the Kortex MCP waitlist. It’s a Kortex Pro feature, at $9.99 a month or $4.92 a month billed yearly. Once your invite lands, you need three things.

1. The Kortex extension, installed from the Chrome Web Store and signed in. Chrome has to stay open while the assistant works. It can sit minimised, and the first call after idle can take up to 30 seconds.
2. Kortex MCP in your assistant. Paste https://mcp.kortex-notebooklm.com/mcp into your client and sign in on the Kortex consent page, or send an API key as Authorization: Bearer kx_mcp_.... In Claude Code it’s one line:

claude mcp add --transport http kortex https://mcp.kortex-notebooklm.com/mcp

3. A way to write to the destination. That’s a Notion connection for Notion, or file access for Obsidian and plain files. The sections below cover each.

The client setup docs cover Claude Code, Claude Desktop and Web, ChatGPT, Cursor, VS Code, Codex CLI and more. If you use an API key, a read-only key covers the reading tools above. However, batch_to_vault saves state, so it needs a key that allows writes, and so does setting up a schedule. Starting from scratch? The sibling guide on using Gemini Notebook in Claude and ChatGPT walks through the first connection.

Notably, item three is the one most readers already have. We analyzed which tools waitlist members said their assistant is connected to, and local files came in at 58%, nearly twice Notion’s 30%.

That gap is why this guide starts with Obsidian; in particular, if your assistant can write files, your vault is already in reach.

How do you import NotebookLM notes into Obsidian?

Point a file-writing assistant at your vault. An Obsidian vault is a folder of Markdown files on your computer. Obsidian “stores your notes as Markdown-formatted plain text files in a vault” and “automatically refreshes your vault to keep up with any external changes” (Obsidian Help, 2026). In other words, whatever writes Markdown into that folder shows up in Obsidian. That makes Obsidian the easier of the two destinations. There’s no second connector to set up and no API to authorise. Claude Code “reads your codebase, edits files, runs commands” (Claude Code docs, 2026), and a vault is a codebase as far as it can tell. Similarly, Codex CLI can “inspect files, make edits” and connect remote MCP servers (OpenAI, 2026).

So open a terminal in your vault folder, start Claude Code or Codex with Kortex MCP connected, and give it this:

You're working in my Obsidian vault. Using Kortex, list the notes in my
"Thesis lit review" notebook and pull them as Markdown with
get_formatted_content. Save each note as its own .md file in
"Inbox/Gemini Notebook/", named after the note title. Add YAML frontmatter
with source_notebook, exported (today's date) and type. Turn any author or
paper names that already exist as notes in the vault into [[wikilinks]].
Don't create or change anything outside that folder.

Each note lands as its own file. With a made-up title and names, Inbox/Gemini Notebook/Sampling methods compared.md looks like this:

---
source_notebook: Thesis lit review
exported: 2026-10-01
type: note
---

Both [[Rivera 2021]] and [[Chen 2023]] use purposive sampling, but only
Chen reports how many interviews it took to reach saturation.

The wikilink step is the part a ZIP export can’t do. The assistant can see your existing notes; consequently, imported material lands already connected to the rest of the vault. Run it once per notebook, and your vault holds every note, linked, without a single copy and paste.

Keep cited answers with batch_to_vault

Notes are what you already saved. Often, however, the value sits in questions you haven’t asked yet. batch_to_vault asks one notebook a batch of questions and keeps “a Markdown answer plus nblm-answer-v1 JSON citation sidecar for each.” It runs in small chunks to avoid timeouts, and you can resume it if something stops halfway.

Use batch_to_vault on my "Competitor pricing" notebook with these questions:
1. What does each competitor charge for its entry plan?
2. Which competitors changed pricing in the last year, and how?
3. What do customers in the sources complain about on price?
Run it in small chunks until it's complete. Then read back each answer and
save it to "Answers/Competitor pricing/" in my vault, one file per question,
with the citations as footnotes at the bottom.

Prefer the raw files? The tool’s download step opens a ZIP Save As dialog in Chrome instead.

How do you connect NotebookLM to Notion?

Connect Notion to the same assistant that has Kortex. In Claude, Notion is a directory connector that lets you “create, edit, search and organize content directly from Claude”, with tools such as create-pages and update-page (Claude, 2026). Add it from Customize, then Connectors (Claude Help Center, 2026). Other clients, meanwhile, reach the same server, https://mcp.notion.com/mcp:

• ChatGPT. Developer mode gives “full Model Context Protocol (MCP) client support for all tools, both read and write” on Pro, Plus, Business, Enterprise and Education accounts on the web (OpenAI, 2026). Kortex already connects there as an MCP app, and Notion says any MCP client can connect to its server, so add Notion next to it.
• Claude Code. Run claude mcp add --transport http notion https://mcp.notion.com/mcp, then sign in from /mcp (Notion Developers, 2026).
• Codex. Add the server under [mcp_servers.notion] in ~/.codex/config.toml and run codex mcp login notion (Notion Developers, 2026).

Now give your assistant the job. This prompt moves a notebook’s notes into Notion, one page each:

Using Kortex, find my notebook "Q3 customer interviews" and list its notes.
Pull them all as Markdown with get_formatted_content.
Then, using Notion, create one page per note under my "Research" page.
Use the note title as the page title, and put the notebook name and
today's date at the top of each page. Show me the list of pages you plan
to create and wait for my OK before writing anything.

Fresh answers work the same way. For instance, ask the notebook a question, then file the answer:

Ask my "Market sizing" notebook: "What growth rates do the sources give for
the EU market, and where do they disagree?" Use ask_question so I get
citations. Then add the answer to my Notion database "Research answers" as a
new row: question as the title, answer in the page body, and each citation
as a bullet with the source name and the quoted passage.

That said, two cautions. Notion’s MCP tools “act with your full Notion permissions” (Notion Help, 2026), so ask for a plan before any writes, as the first prompt does. And very long, heavily formatted Markdown can fail in one call with a validation_error (Notion Developers, 2026). If a 20-page briefing doc bounces, ask for one page per section.

What about Google Docs, Sheets and plain files?

Use the export Google already built, through Kortex. export_artifact can call Gemini Notebook’s native Drive export for a Studio output. Google’s help center describes that export for notes: notes “populate in a new Google Doc.” Data tables go to Sheets, with “each data table” in “a separate tab” (Gemini Notebook Help, 2026). Your assistant triggers it without you opening the notebook.

Alternatively, answers and notes can reach Docs another way. Google Docs has imported Markdown since July 2024, from File, then Open, or “Open with Google Docs” in Drive (Google Workspace Updates, 2024). Have your assistant save the Markdown file, then open it in Docs.

Studio outputs need the right tool for each format. This prompt sorts them:

In my "Biology 201" notebook, list the Studio artifacts and their status.
For each completed one:
- report or study guide: export it as Markdown and save it to ./exports/
- data table: export it to Google Sheets and give me the link
- flashcard deck or quiz: read it with get_artifact_content and save it as
  a Markdown file of question / answer pairs
- audio or video: give me the media link
Skip anything still in progress, and tell me what you couldn't export.

How do you keep it running every week?

Give the export a schedule on each side. A Kortex schedule is an automation that runs hourly, daily or weekly inside the extension while Chrome is open. It can ask a notebook and save the answer (Kortex MCP docs, 2026). Claude Code Desktop’s local scheduled tasks run “on your machine with direct access to your files and tools” (Claude Code docs, 2026). However, Kortex schedules can’t write to Notion or your vault on their own. They live in the extension, and only your AI client can reach outside apps. Therefore, pair them:

1. Kortex fills the notebook. A schedule asks your project notebook a standing question every Monday and saves the answer as a note. The daily briefing guide shows how to set one up.
2. Your assistant carries it out. A local scheduled task in Claude Code Desktop, set to Weekly, runs a prompt like the one below against your vault folder.

Weekly export. Using Kortex, list the notes in my notebooks "Project Atlas"
and "Thesis lit review". Compare their ids with the list in
".kortex-export-log.md" in this folder. For every note that isn't in the
log, pull it as Markdown with get_formatted_content, save it to
"Inbox/Gemini Notebook/<notebook name>/", then append its id to the log.
If nothing is new, write nothing and say so.

The log file matters. list_notes returns ids and titles, not dates; hence the log is how the task knows what it already moved. One honest catch: the Desktop task “only fires while the app is open and your computer is awake,” and Kortex needs Chrome open too. A closed laptop runs nothing on either side.

Additionally, if you run the same multi-step job by hand often, save the Kortex half as a workflow. The AI workflows guide shows how create_workflow and run_workflow chain Kortex tools into one command.

What are the limits?

The big one is direction: this is a one-way carry, not a live sync. Of the 776 waitlist respondents, 46 asked to keep their notebooks in live sync with other tools (Kortex MCP waitlist survey, 2026). Automatic two-way sync isn’t something Kortex MCP does today. Edit a page in Notion, in other words, and the notebook won’t change.

A few more to plan around before your first run:

• Google’s usage limits still apply. Every ask_question and batch_to_vault question counts against your Gemini Notebook quota. That quota has refreshed every five hours since September 2, 2026, and our usage limits guide explains how far it stretches.
• Reading saved notes is light. get_formatted_content and list_notes read what’s already there. They don’t spend Kortex export credits or ask Gemini Notebook anything new.
• Media stays a link. Audio and video come back as a direct media link, not as text your assistant can file.
• Writes need care. Use a read-only Kortex key when the assistant only reads, and keep your client’s write confirmations on for Notion. The security docs cover key profiles.

---

Getting notes out of Gemini Notebook stops being a chore once your assistant does the carrying. Kortex reads the notebook, and Notion, Obsidian or Docs gets the result in the shape you want. If that’s the weekly job you’ve been doing by hand, this is the place to hand it off.

Frequently asked questions

Does NotebookLM connect to Notion?

Not on its own. Gemini Notebook (formerly NotebookLM) has no Notion source type and no public consumer API, so it can’t read or write a Notion workspace. Notion’s MCP server connects AI apps like Claude, ChatGPT and Cursor to Notion, so your assistant can carry notes over with Kortex MCP, a private beta for invited Kortex Pro members.

Can I sync NotebookLM to Notion automatically?

Not as a live two-way sync. Your assistant carries notes over when you ask, through Kortex MCP on the reading side and its own Notion connector on the writing side. For a hands-off routine, a Claude Code Desktop scheduled task can repeat the job daily or weekly while your computer is awake and Chrome is open.

How do I get NotebookLM notes into Obsidian?

An Obsidian vault is a folder of Markdown files, so any assistant that can write files can fill it. Open Claude Code or Codex in the vault folder, connect Kortex MCP, and ask it to pull your notes as Markdown with get_formatted_content and save each one as a .md file. Obsidian picks up the new files on its own.

Does Kortex write to Notion itself?

No. Kortex MCP reads and writes inside Gemini Notebook only. The writing on the Notion side is done by your assistant’s own Notion connection, such as Claude’s Notion connector or Notion’s MCP server added in ChatGPT developer mode, Claude Code or Codex. Kortex never reads or touches your Notion workspace.

Can I export Audio Overviews, flashcards and data tables this way?

Yes, with different tools. export_artifact returns a report as Markdown, a data table as CSV, or a direct media link for audio and video, and can send outputs to Google Docs or Sheets. get_artifact_content reads the actual cards in a flashcard deck, the questions in a quiz, or the tree of a mind map.

Do I need Kortex Pro for this?

Yes. Kortex MCP is a Kortex Pro feature, at $9.99 a month or $4.92 a month billed yearly, and it’s in private beta, so only invited members can connect. Invites go out from the Kortex MCP waitlist. Manual export buttons are on the free plan, with 10 exports a day.

---

Editorial note: fact-checked against the official help pages and docs linked above on October 1, 2026. Spotted something out of date? Contact us and we will fix it, or read more about Kortex.