notion-site-fetch 0.1.0

Fetch any public Notion page and print it as Markdown — no token, no headless browser.

notion-site-fetch

Fetch any public Notion page and print it as Markdown to stdout. No API token. No login. No headless browser. Just plain HTTPS calls to the endpoints Notion's own web client uses.

If the page opens for anyone in an anonymous browser tab, this tool can read it.

Install / run

The package is published on PyPI. The fastest way to run it is with uvx, which fetches and executes the tool on demand:

uvx notion-site-fetch https://example.notion.site/ > page.md

You can also install it permanently:

uv tool install notion-site-fetch
notion-site-fetch https://example.notion.site/ > page.md
# or with pipx:
pipx install notion-site-fetch

Requires Python 3.13+.

Usage

notion-site-fetch <url>

Markdown is written to stdout. Use shell redirection to save it:

notion-site-fetch https://example.notion.site/             > page.md
notion-site-fetch https://example.notion.site/Some-Subpage >> notes.md

Errors go to stderr; the exit code is non-zero on failure.

Accepted URL forms

URL	What it fetches
https://<sub>.notion.site/	Site's public home page
https://<sub>.notion.site/<slug-or-id>	A specific page on the public site
https://www.notion.so/<...>-<32-char-page-id>	A notion.so page URL

Any URL whose path ends in a 32-character Notion page id is accepted even without a .notion.site host.

Behavior

• Sub-pages stay as links. Only the page you ask for is fetched. To expand a linked sub-page, run the tool again with its URL.
• Toggles/dropdowns are expanded inline. Nothing is hidden behind a fold; nested toggles are followed recursively.
• Private pages fail clearly. If the page isn't publicly readable, the tool exits with an error rather than silently returning nothing.
• Pagination is handled for long pages.

What gets rendered

Headings (H1–H4), paragraphs, bold/italic/strikethrough/code inline formatting, links, bulleted/numbered/to-do/toggle lists with nesting, quotes, callouts (as blockquotes), dividers, code blocks (with language), images, bookmarks, video/file/PDF/audio embeds (as links), and inline equations.

Things that are deliberately skipped or simplified: table-of-contents and breadcrumb blocks (Markdown renderers generate their own), column-layout wrappers (children are flattened into normal flow), and collection-view databases (only the title is emitted — Notion does not expose row-level data to anonymous viewers).

How it works

Notion's public sites are served as an empty React shell and hydrated by client-side calls to three internal endpoints:

1. POST /api/v3/getPublicPageData with {"spaceDomain": "<sub>"} resolves a *.notion.site subdomain to a spaceId.
2. POST /api/v3/getPublicSpaceData with that spaceId returns the publicHomePage block id — the root page of the site.
3. POST /api/v3/loadCachedPageChunkV2 walks the page block by block, paginated with a cursor. Toggle children that the chunk loader skips are pulled in afterwards via syncRecordValues.

All endpoints are unauthenticated and return JSON. The tool issues these requests against the site's own subdomain (*.notion.site) where possible to avoid Notion's cross-cell routing errors on www.notion.so.

License

MIT — see LICENSE.