You may have heard of Programmatic Tool Calling from anthropic or Code Mode from cloudflare for calling mcp servers. I have been working on an open source version of this and recently put together some basic docs, and thought this page would be a good one to share.

There are plenty of other features to explore, but this one alone has resulted in significant improvements in the speed of execution in many tasks including minecraft playing agents, computer use agents, and for querying and aggregating information from my telemetry systems in a lightweight low risk way.

if i didn't have to prompt it to learn from its mistakes and it just "intuitively" knew to do that

have built in this space which led me to develop a minizinc mcp server [0] for scheduling bocce tournaments [1]. scheduling with constraints is a np hard problem and it makes sense people struggle. tools exist to solve this problem but they are complex and hard to use for non technical folks, and even technical folks. am hoping a tool like this can bridge the gap and would like to bring it to your awareness if you aren't already thinking about the problem this way :)

edit: after reading a bit more of description looks like yall are taking a similar approach, kudos!

[0] https://github.com/r33drichards/minizinc-mcp

[1] https://github.com/r33drichards/bocce-scheduler

This is awesome! Completely agree: modeling each real life scenario as a constraint satisfaction problem is tricky in and of itself (especially with the diversity of non-intersecting constraints we encounter) and something we are actively working on. Using LLMs as a layer above has made it much more tractable. Curious how the bocce scheduling has fared in real world scenarios. How was the performance?

yep! thats the motivation behind https://github.com/r33drichards/mcp-js

I want to be able to give agents access to computation in a secure way without giving them full access to a computer

Does this support bedrock?

Could add support if you need it! Just let me know :D

Is it possible to provide a llm a skill through the mcp resource feature?

In a way yes, you can reduce context usage by a non-negligible amount approaching it this way. I'm investigate this on my skill validation/analysis/bidirectional MCP server project and hope to have it as a released feature soon: https://github.com/athola/skrills

It’s also possible to implement an MCP as a skill

as an infrastructure engineer the idea of being able to train computer use agents without provisioning infrastructure sounds amazing!

a common use case i run into is i want to be able to configure corporate vpn software on windows machines. is there a link for a getting started guide i could try this out with?

Yes, in a simulated environment you can do this today using plain JS and connecting to a real VPN, while driving the desktop UI. No infra provisioning needed.

If you need a real Windows OS + corporate VPN, we also support binding agents to actual Windows sandboxes. This example shows automating a Windows app behind a VPN: https://cua.ai/docs/example-usecases/windows-app-behind-vpn

you'll need to define a new task in the cua-bench registry first though - just sign up on the website for early access!

have used it, and i do like it, but the licensing situation is not great. It open source but its not free software by any means.

wrote about this a bit too in https://www.robw.fyi/2025/10/24/simple-control-flow-for-auto...

ran into this when writing agents to fix unit tests. often times they would just give up early so i started writing the verifiers directly into the agent's control flow and this produced much more reliable results. i believe claude code has hooks that do something similar as well.

clearly you have never worked in enterprise