Skip to main content

Connect Gemini Notebook to Zapier using a Webhooks step

· 4 min read
Romain Peyrichou
Romain Peyrichou
Author of NotebookLM MCP

Search Zapier's app directory for NotebookLM and you find a page that says the integration has not been built. There is no trigger, no action, no module. That is still true in late 2026.

You do not need one. Zapier ships a generic HTTP action — Webhooks by Zapier — and your Gemini Notebook server already speaks plain HTTP with a key. Point one at the other and a Zap can ask a notebook a question and read the citation-backed answer like any other step.

A note on the name. Google renamed NotebookLM to Gemini Notebook in July 2026. Same product, old links redirect, and everything below applies to both — the two names are used interchangeably here because that is how people still search for it.

Why is there no Zapier app?​

A Zapier app needs an API with a key, and the consumer Gemini Notebook has neither. Google's documented API is the enterprise one, behind a Cloud project and an administrator — an app built on it would serve only the users who could already script it.

So the piece Zapier is missing is the thing our server adds: a small HTTP endpoint that holds your Google session and exposes the operations as ordinary routes. Once that exists, the "no app" problem is a non-problem, because the generic Webhooks action reaches any HTTP endpoint.

What is the recipe?​

One step: Webhooks by Zapier → Custom Request.

  • Method: POST
  • URL: https://your-host/ask
  • Data (JSON body):
{
  "question": "What changed in this week's board pack?",
  "notebook_id": "<your-notebook-id>",
  "source_format": "markdown"
}
  • Headers:
X-API-Key: your-token
Content-Type: application/json

Set the Data Pass-Through? field to false so Zapier sends the JSON body above rather than the trigger's fields. The server also accepts Authorization: Bearer your-token if you prefer the standard header.

The one field people forget is source_format. It defaults to none, and with none the answer comes back without citations. Set it to markdown (or text) and the sources arrive with the answer.

What does Zapier get back?​

A JSON object you can map straight into the next step:

{
  "success": true,
  "data": {
    "status": "success",
    "answer": "The pack adds a cash-runway slide and revises Q4 guidance…",
    "session_id": "…",
    "sources": {
      "format": "markdown",
      "citations": [
        {
          "marker": "[1]",
          "number": 1,
          "sourceText": "Runway extended to 14 months…",
          "sourceName": "Q3 board pack.pdf"
        }
      ],
      "extraction_success": true
    }
  }
}

In the Zap's later steps, data__answer is the text, and each entry under data__sources__citations carries a marker, a number, the quoted sourceText and, when known, the sourceName. Keep data__session_id if a follow-up step should continue the same thread instead of starting fresh.

The answer is just data from there: it can land in a sheet row, a Slack message, a database, or a document, like anything else in a Zap.

Where does the endpoint have to run?​

Somewhere Zapier's cloud can reach — that is the real constraint, and it is the same one the n8n, Zapier and Make overview describes.

Zapier runs in Google's cloud, not on your laptop, so the server has to sit on a host Zapier can see. The moment it does, you are exposing an endpoint that holds a live Google session, so set a token: our server refuses to start on a public interface without one, which is deliberate — the failure is louder than a leak.

When is this the wrong tool?​

  • When latency matters. An answer takes seconds, not milliseconds. Do not put it in front of someone waiting on a synchronous Zap.
  • When the Zap must never stall. Usage is a compute allowance that refreshes every few hours and can run out mid-run.
  • When exposing the server to the public internet is the only way to reach Zapier. Weigh that against keeping the work on a self-hosted automation tool on the same machine instead.
  • When a plain Gemini API call would do. If you do not need grounding in your own documents, you do not need any of this.

Trying it​

Run the server, then send the exact request your Zap will send:

npx -y -p @roomi-fields/notebooklm-mcp notebooklm-mcp-setup-auth
NOTEBOOKLM_API_KEY=your-token HTTP_PORT=3000 node dist/http-wrapper.js

curl -s -X POST http://localhost:3000/ask \
  -H "X-API-Key: your-token" \
  -H "Content-Type: application/json" \
  -d '{"question":"What changed this week?","notebook_id":"<id>","source_format":"markdown"}'

Node 22 or newer. The full endpoint list is in the API reference, and if you self-host your automation the n8n guide wires the same call as a native node.