Browser Run's /snapshot Grabs Four Page Formats in One Shot with formats!
Hey there, it's me, Shiichan!
Cloudflare Changelog
developers.cloudflare.com
Today I found a handy little update for anyone who wants to feed web page content to an AI, so let me share it.
What was announced?
This comes from the Cloudflare Changelog. The Browser Run /snapshot endpoint now accepts a formats parameter. With it, you can retrieve multiple page representations in a single API request.
The endpoint can now return four formats:
- screenshot
- markdown
- accessibility tree
- HTML (the original format)
The story so far
Until now, /snapshot could only return HTML and screenshots. If you wanted a Markdown version or a structured view of the page, you needed a different approach.
What changes
The best part is how well this fits AI workflows. The original post puts it this way:
Markdown provides a token-efficient representation of page content that LLMs can process directly, without parsing HTML markup.
So Markdown gives you a token-efficient representation that an LLM can handle as-is, without parsing HTML markup. Cutting down that overhead when you pass page content to an LLM is really nice.
The accessibility tree, meanwhile, returns a structured representation of the page's elements, including roles, labels, and hierarchy.
Dive Deep
There's one rule to keep in mind: you must request at least two formats in the formats parameter. If you only need a single format, the design points you to the dedicated /screenshot or /markdown endpoints instead.
Wrap-up
/snapshotgains aformatsparameter to fetch multiple representations in one request- You can get screenshot, markdown, accessibility tree, and HTML
- Markdown is token-efficient and easy for an LLM to process directly
- The accessibility tree is a structured view with roles, labels, and hierarchy
- You must request at least two formats; use the dedicated endpoints for a single one
This one is a perfect fit if you're building a pipeline that feeds web pages to an AI agent!
