Markdown for Agents Now Converts Up to 6 MiB with an In-Process Streaming Engine

Hi, it's Shii-chan! Today I found an update that makes Markdown conversion for AI agents lighter, so let me walk you through it.

What was announced?

The Cloudflare Changelog has an update to Markdown for Agents.

The conversion engine now runs in-process at the edge and streams. It processes HTML as it arrives, which reduces conversion overhead and memory use.

Until now, the HTML response was buffered in full and then sent to a separate conversion service to be turned into Markdown.

That meant waiting for everything to arrive and then handing it off, which cost extra overhead and memory.

Conversion now happens in the same process, as a stream.

Longer documentation pages should now fit within the limit more often.

Here are the limit and the response header changes. The key point is that some headers are gone.

The 6 MiB limit applies to decompressed HTML, not to the size of the compressed response
Converted responses no longer include the x-markdown-tokens and x-original-tokens headers
Content-Length is removed rather than recalculated, because the Markdown body is streamed

Depending on your client, you may need to adjust:

See the Markdown for Agents documentation for the details.

Conversion moved from a separate service to an in-process streaming engine at the edge
The decompressed HTML limit goes from 2 MiB to 6 MiB
x-markdown-tokens, x-original-tokens, and Content-Length are no longer returned
Clients that use token counts need to compute them on their own

If you serve content to agents with Markdown for Agents, or build clients that read these headers, this one is worth a look!