Added a read counter to posts on my blog. Can now see exactly how many real people vs bots read my stuff. #ai #bots
| Personal site | https://nmm.ee |
| Open Source Work | https://git.nmm.ee |
Asko Nõmm
@iamnmm
- 7 Followers
- 11 Following
- 34 Posts
Senior Software Engineer at https://codescene.com. I work on a whole bunch of AI tooling to, hopefully, make your life easier.
TL;DR of it is that for input tokens, they seem to have no difference, but then the main difference comes when using manually created CLI integrations with SKILL’s, where it differs behaviourly, both with their own pros and cons.
A lot of people seem convinced that MCP’s are bad when it comes to token spend, or that MCP’s are dead, and that manually created CLI integrations are the only true way. I set out to find out the truth, and so I ran a little experiment and did a bunch of digging on how things really work under the hood.
Spoiler: MCP’s are not dead.
Do MCP’s use more tokens than CLI’s?
There’s a long-held belief out there that MCP’s are bad, and that you should actually use CLI’s instead, if you want to save on token spend when it comes to LLM tool use. To me that argument never really made any sense, because you still have to provide the LLM context so it knows what CLI to run, how to run it and how to interpret the result of it. Things that the MCP does for you. How come the CLI takes fewer tokens?
KDEOn #IloveFS day, "This week in Plasma" brings the news that...
Plasma 6.6 is nearly ready for shipping! Look out for it next Tuesday.
... And videos in Plasma 6.7's SDDM login screens can be previewed in System Settings, KWin dialogs have been overhauled and polished, and much more.
https://blogs.kde.org/2026/02/14/this-week-in-plasma-finalizing-6.6/
You know what's crazy to me? I paid tons of money for a 16" M4 Max 48GB #macbook pro, because its processing power is about 2x compared to my #Framework 16's ryzen ai 9 hx 370 .. and yet, the user experience of #macos is noticeably slower compared to my @fedora @kde. A lot slower. The animations take forever when opening/closing apps, and it being Apple, you of course can't change any of that.
2x faster machine on paper, 2x slower machine in day to day use. Make it make sense.
It has become evident to me that the only viable way to test a MCP server is via integration tests, not via unit tests. #mcp
"This Week in Plasma" brings the news that Plasma 6.6 lands in little over two weeks, choosing mixed skin tones for emojis gets a friendly pop-up, and you can set a global keyboard shortcut to clear the notification history, among many other marvelous things.
https://blogs.kde.org/2026/01/31/this-week-in-plasma-getting-6.6-ready-for-release/
I just put an order in for the Ryzen AI 9 HX 370 Mainboard for my #framework 16 laptop. Personally hoping it's a bit more efficient with energy and that I gain a bit of battery life. Some people seem to have 3-5W on low use, which would be quite amazing to see.
Mike Clarke📣 Available for new contracts 📣
I'm available for new contracts as senior developer, tech
lead, or head of engineering.
Technically, I build scalable, reliable APIs, back-end, and systems software with #python / #elixir / #rustlang / #clojure or similar.
As a lead, I build high-ownership, high-impact technical teams.
Recent focus has been on regtech, especially in the legal and health fields.
Please do get in touch for a discussion if you have anything that might be a fit!