> Every MCP server injects its full tool schemas into context on every turn
I consider this a bug. I'm sure the chat clients will fix this soon enough.
Something like: on each turn, a subagent searches available MCP tools for anything relevant. Usually, nothing helpful will be found and the regular chat continues without any MCP context added.
I'll add to your comment that it isn't a bug of MCP itself. MCP doesn't specify what the LLM sees. It's a bug of the MCP client.
In my toy chatbot, I implement MCP as pseudo-python for the LLM, dropping typing info, and giving the tool infos as abruptly as possible, just a line - function_name(mandatory arg1 name, mandatory arg2 name): Description
(I don't recommend doing that, it's largely obsolete, my point is simply that you feed the LLM whatever you want, MCP doesn't mandate anything. tbh it doesn't even mandate that it feeds into a LLM, hence the MCP CLIs)
Yup, routing is key. Just like how we've had RAG so we don't have to add every biz doc to the context.
I agree with the general idea that models are better trained to use popular cli tools like directory navigation etc, but outside of ls and ps etc the difference isn't really there, new clis are just as confusing to the model as new mcps.
I don’t think so. Without a list of tools in context the ai can’t even know what options it has, so a RAG like search doesn’t feel like it would be anywhere near as accurate
Interesting pricing differential. Seems in your country, that IdeaPad is significantly cheaper than the price in the US. But for your Macbook Neo, it's the other way around.
No idea. Maybe Lenovo includes purchasing power in the price calculation for some reason, such as making more money in the U.S. while gaining market share here in Czechia, where purchasing power is lower. Apple may be able to afford not to do that.
But qwen3.5 35b is worse than even Claude Haiku 4.5. You could switch your Claude Code to use Haiku and never hit rate limits. Also gets similar 50tps.
I haven't tried 4.5 haiku much, but i was not impressed with previous haiku versions.
My goto proprietary model in copilot for general tasks is gemini 3 flash which is priced the same as haiku.
The qwen model is in my experience close to gemini 3 flash, but gemini flash is still better.
Maybe it's somewhat related to what we're using them for. In my case I'm mostly using llms to code Lua. One case is a typed luajit language and the other is a 3d luajit framework written entirely in luajit.
I forgot exactly how many tps i get with qwen, but with glm 4.7 flash which is really good (to be local) gets me 120tps and a 120k context.
Don't get me wrong, proprietary models are superior, but local models are getting really good AND useful for a lot of real work.
But in your country (Spain), Telefónica de España laid off 3649 workers in Dec 2023 (about 40% of that unit) despite growing net income by 17% that year.
Nice googling, but that’s just an example that proves my point.
They had to go through a process extensively justifying losses (mostly that certain jobs were no longer relevant as they were pre-digital workforce), negotiate with unions and offer voluntary leaving conditions.
The resulting offer was good enough that more workers applied to be fired than were necesssary. For context, the deal was basically to pay them 70% of their current salary from the dismissal moment until their retirement at 63.
I can, easily. Speaking for my country and assuming you are not on a fixed-term contract, you can only get fired for one of the three reasons:
1. Company's in financial trouble and forced to downsize.
2. The position becomes obsolete and there's no option to transition you into some other role. In this case, the company can't hire anyone with a similar-enough skillset to yours for at least a year (or maybe even longer, I'm not sure).
3. Gross incompetence, in which case you need to be given an opportunity to course correct via a few documented warnings before being fired. Every warning requires your signature so that the company can't just make them up and backdate them.
That said, you don't become a permanent employee on day one, a company can issue up to three fixed-term contract before being forced to give you a permanent one. If you're on a fixed-term contract, they can just not extend it without having to satisfy any of the criteria above. But after a maximum of 3 years at the same company (as the maximum length for a fixed-term contract is one year), the criteria for firing you increases drastically.
So, the only way this could happen in my country is if the company stops renewing fixed-term contracts for recent employees, but then it wouldn't all be at the same time and you'd get the hint before the time comes to plan accordingly.
People don't want to leak their personal information, duh. The country they are from is another bit of information that you can't take back from the internet once published. Why would you do that?
I think that barrier may be weakening. I reckon that the people most concerned about crime are willing to sacrifice their privacy and defer to their government to prevent crime.
But I think the shenanigans of ICE are making people more aware of the importance of privacy. Look at the backlash Ring (lost dog superbowl ad) and Discord (age verification) and Nest (Guthrie case) received just this year.
The people I have in mind are the rich, poor and those who fantasize becoming wealthy and fear going broke. I’m uncertain how much these demographics account for the US population and empirically speaking I’m unsure of the gravity of the PR stirs you named. I really don’t know if privacy is the foremost concern when the types of people I’m thinking of consider ICE either.
It really seems to boil down to whether these types of people can be effectively sold on the virtue of tearing down the barriers of privacy and government. If they aren’t already implicitly sold to that then all it takes is for the powers that be to do a better job at marketing their initiative.
I consider this a bug. I'm sure the chat clients will fix this soon enough.
Something like: on each turn, a subagent searches available MCP tools for anything relevant. Usually, nothing helpful will be found and the regular chat continues without any MCP context added.
reply