well if you read this you will know that progressive disclosure is actually now possible with MCP... well it always has been as a pattern in the spec - but claude code actually honours the spec as of Jan 2026... so there is no bloat baby
Yeah, I'm well aware of the pattern, and I'm aware of the Cloudflare code paper and I know that the overhead of MCP is still unneeded because you still have to have those open HTTP connections or local connections or whatever the case is. Those are completely unnecessary when your harness can use a tool like a CLI. It's way more efficient. You can explain it in less tokens or you can type script out the entire primitive so that you don't need the overhead.
I still don't see it. I appreciate that you do and that you made a product to deal with it but I just think it's still a bad pattern.
tbh the whole code mode thing is actually kind of what were doing with commandable mcp.
The agent is essentially given a sandbox to write abitrary code against an API, with the users credentials for that api stored securly...
The pattern i assumed is that the agent would then 'save' this code as a tool to reuse later...
But youve got me thinking - maybe commandable should be more just like full code mode every time - every tool call be a new custom script written right then and executed immediately.
Maybe 'saving' those actions for later is just something the agent could do if it needs it more frequently...
"we do it better-er"
Mcp is dead, gtf out of my context, progressive disclosure is a superior system, server connection bloat was a shittastic idea
well if you read this you will know that progressive disclosure is actually now possible with MCP... well it always has been as a pattern in the spec - but claude code actually honours the spec as of Jan 2026... so there is no bloat baby
Yeah, I'm well aware of the pattern, and I'm aware of the Cloudflare code paper and I know that the overhead of MCP is still unneeded because you still have to have those open HTTP connections or local connections or whatever the case is. Those are completely unnecessary when your harness can use a tool like a CLI. It's way more efficient. You can explain it in less tokens or you can type script out the entire primitive so that you don't need the overhead.
I still don't see it. I appreciate that you do and that you made a product to deal with it but I just think it's still a bad pattern.
tbh the whole code mode thing is actually kind of what were doing with commandable mcp.
The agent is essentially given a sandbox to write abitrary code against an API, with the users credentials for that api stored securly...
The pattern i assumed is that the agent would then 'save' this code as a tool to reuse later...
But youve got me thinking - maybe commandable should be more just like full code mode every time - every tool call be a new custom script written right then and executed immediately.
Maybe 'saving' those actions for later is just something the agent could do if it needs it more frequently...
hmmmmmm food for thought
legit