Jared

Cloudflare's new code mode technique.

MCP is, for now at least, the standard way AI agents use external tools.

But it sure does fill up the model's context window with a lot of cruft.

To combat this, Cloudflare came up with code mode, which rhymes, so you know it's good.

Quote, code mode is a technique we first introduced for reducing context window usage during agent tool use.

Instead of describing every operation as a separate tool, let the model write code against a typed SDK and execute the code safely in a dynamic worker loader.

The code acts as a compact plan.

The model can explore tool operations, compose multiple calls, and return just the data it needs."

As a result of this, Cloudflare created a new MCP server for their entire API.

With just two tools, search and execute, the server is able to provide access to the entire Cloudflare API over MCP while consuming only around 1,000 tokens.

The footprint stays fixed no matter how many API endpoints exist.

End quote.

This is a good example of what I've been talking about a lot recently.

The models are getting marginally better, but the traditional software engineering around the models squeeze out huge wins by equipping them better and better.

The only moat left is money.

Here's Elliot Bonneville.

Quote, every morning, a few thousand people wake up and ship something.

A tool, a SaaS, a newsletter, an app that does the thing the other app does, but slightly differently.

They post it on Hacker News.

Nobody clicks.