Sulat.com / Writing on AISaturday, November 21, 2026
Agents/21 November 2026

Google’s Gemini Notebook API now takes file uploads, but the MCP server can’t send them

Google’s Gemini Notebook API can now ingest PDFs, Word files, audio and images through a REST upload route. The MCP server has no file tool, and the API still can’t delete sources.

Glossy 3D Gemini Notebook arches with a PDF page, a photo and an audio waveform card sliding into them on a dark blue stage.

At a glance

  • Google's Gemini Notebook API now accepts uploaded files, including PDFs, Word and PowerPoint files, EPUBs, CSVs, audio and images, through a REST route, up to 200 MiB and 500,000 words each.
  • Seth Ladd announced the change on X; Google's MCP page lists seven tools and none of them takes a file, so an agent on MCP alone can't upload.
  • The API has no delete method and no way to ask a notebook questions, and it works only with personal Google accounts during the free preview.

Google’s Gemini Notebook API now takes file uploads. A new upload route accepts PDFs, Word and PowerPoint files, EPUBs, CSVs, audio and images, none of which the API could ingest when its preview opened. Seth Ladd, who carries a Gemini Notebook label on X, announced it on 21 November NZ time.

Seth Ladd on X, 20 November US time: files can now be saved into Gemini Notebook through the public preview API.

What the new route takes

Google’s sources guide, last updated on 20 November, documents the upload endpoint as a POST to notebook.googleapis.com/upload/v1/{notebook}/sources:upload. Each upload creates one source. Processing runs in the background, so a client polls the source until it reads ACTIVE or FAILED.

Before and after: four source types in the October preview, plus seven uploaded file kinds now, with a 200 MiB file limit.
Uploaded files join web pages, YouTube links, text and Markdown as source types.

The request’s Content-Type has to be one of the types in the table. Anything else, including Excel spreadsheets, gets a 400 error.

File typeContent-Type
PDFapplication/pdf
Word, PowerPoint, OpenDocument TextDOCX, PPTX and ODT
Rich Text Format, Hangul Word ProcessorRTF and HWP
EPUBapplication/epub+zip
CSV, plain text, Markdowntext/csv, text/plain, text/markdown
Audioany audio type
Video3GPP and 3GPP2 only
Imageany image type
File types the Gemini Notebook API accepts for upload, from Google’s sources guide, checked on 21 November 2026.

A file can be up to 200 MiB and 500,000 words. Each user can start 20 uploads a minute, and going over returns a 429 error with a retry delay. A notebook’s source allowance depends on the plan, according to Google’s rate limits page: 50 on Free, 100 on Plus, 300 on Pro and 600 on Ultra.

Three limit cards: 200 MiB per file, 500,000 words per source, 20 uploads a minute per user, plus source caps by plan tier.
Three numbers cap what an agent can upload, and the plan tier caps how many sources a notebook holds.
See the simple upload request

This is Google’s own example for a PDF. Without the displayName parameter the source is called Untitled.pdf.

Code
curl -s -X POST \
  "https://notebook.googleapis.com/upload/v1/${NOTEBOOK_NAME}/sources:upload?uploadType=media&source.displayName=Quarterly%20report" \
  -H "Authorization: Bearer $ACCESS_TOKEN" \
  -H "x-goog-user-project: $PROJECT_ID" \
  -H "Content-Type: application/pdf" \
  --data-binary "@report.pdf"

The MCP server still can’t send a file

Ladd’s post tells developers to connect an agent to the MCP server or use the API directly to create notebooks, add text and now add files. Google’s MCP page lists seven tools: create, get, list and update for notebooks, and create, get and list for sources. None of them takes a file, and the page says uploads go through the REST endpoint instead.

An agent that only has the MCP connection can add a URL, text or Markdown, and it can’t add a PDF. To upload, the agent needs a shell or code to call the REST route with its own access token.

REST API accepts uploads, MCP server doesn’t; roadmap still lists delete, ask a notebook, make outputs and Drive links.
Uploads go through REST only; an agent on the MCP server needs code or a shell to send a file.

What changed since the preview opened

Gemini Notebook is the former NotebookLM, renamed on 16 July. When the API preview was covered on 8 November, binary uploads sat on Google’s roadmap, and the sources page dated 29 October said so. The roadmap page, updated on 19 November, no longer lists them.

Five items remain on it: Google Workspace accounts, deleting notebooks and sources, Google Drive URLs as sources, retrieving information from a notebook and generating artifacts. Google calls the roadmap informational and subject to change, and it gives no dates.

What an agent still can’t do

The REST reference lists create, get, list and patch for notebooks and create, get and list for sources. There is no delete, so a script that uploads in a loop fills a notebook’s source allowance and can’t clear it. There is also no way to ask a notebook a question through the API, which means answers still come from the Gemini Notebook apps.

The preview is free and limited to personal Google accounts. Workspace accounts get a 403 error, and an account that has never signed in to Gemini Notebook gets a 404 when it creates a notebook. Google’s Python client library can upload with resumable=False and no request body, but it names every source Untitled with the file’s extension, and its multipart and resumable uploads fail with HTTP 400. Naming a source needs the raw REST request.

Gemini Notebook Enterprise, a separate Google Cloud product, documents its own uploadFile method on a v1alpha endpoint. Google takes feature requests for the personal API in its Issue Tracker.


Discover more from Sulat AI & Tech News

Subscribe to get the latest posts sent to your email.