Bin the MCPs. We are stripping context bloat and slashing costs by letting AI read a local code folder, not a mountain of prompts. We chat Anthropic's make it dumber approach, show how to plug repos into a project, and ask for your takes for next week's readout on Command AI.
In this video, we discuss the drawbacks of using MCPs in AI workflows and explore practical ways to reduce reliance on them. We explain how replacing large MCP bundles with lighter code folders can significantly cut context usage, lower token consumption, and reduce overall processing time and cost. We also reference a recent Anthropic article that outlines practical methods for implementing this approach. This video continues last week’s discussion and offers actionable insights for developers and AI enthusiasts.
------------------------------------------------------------
0:00 - Intro
0:23 - Explaining Why MCPs Are a Problem
1:27 - How to Replace MCPs + Article Reference
------------------------------------------------------------
Discussion in this weeks news in AI, web and design with @Kabarza and @0x5am5
Unlock the full potential of your online presence with Kabarza and Samuel—experts in web design and development (respectively), powered by cutting-edge AI solutions. We blend creative design with advanced tech to deliver smart, high-impact websites that stand out. Ready to elevate your business? Contact us today and see what AI-driven innovation can do for you!
Kabarza: https://kabarza.com
Samuel: https://samuelgregory.co.uk
🎙️ New to streaming or looking to level up? Check out StreamYard and get $10 discount! 😍 https://streamyard.com/pal/d/4597668583964672
#AI #Anthropic #MCP
#ai #podcast #aidesign #aidevelopment #vibecoding #webdesign #webdevelopment #ainews #webnews #designnews #devnews
Transcript
So, this is the comments section where we take a look at what you have to say. Uh, let us know what you think about anything you hear down in the comments and we might just read it out. I have a less of a comment, more of a follow on from something we spoke about last week. So, this is our video making um AI dumber. It's it's an article from Anthropic and they they talk about this pattern. You can see there's a servers folder. I don't want to move too much because we'll lose the There we go. there's a service folder because AI is really good at writing code. So rather than giving an MCP loaded into say claw code, give it a folder of code that interacts with Google Drive and and Salesforce here and then AI will be smart enough to figure out, you know, what code it needs to write to import those libraries. And they found an 150,000 tokens got reduced to 2,000 tokens of context at 98.7% reduction time and costs saving because they're not using MCPs. And I was left at the end of that being like what how does this how does this all tie together? And that's where this article comes in where he talks about um how to actually do it. And ultimately, here we go. There's some there's some um bash stuff here. Ultimately, the way I interpret it is you can actually download the or you know if the if the MCP is on on a GitHub repo, you basically download the GitHub repo into the your folder of your project and you say in your rules, you basically say there is a there is a there is a folder here for you to interact with the Google Drive MCP and whatever. Uh, granted I've not read this in in much depth, but I watched a kind of review on it and this is like again just further enforcing or a you know a something to read when it comes to trying to reduce MCP users because MCPs are they just they just they just suck. They they really do take up so much context. Some things you can't get around. I've tried, but um yeah, it's definitely something I'm if if I get some downtime, I'm definitely going to explore kind of removing MCPS from my workflow as much as I can and use this article to actually do this. So, and I'm looking for the dates. So, November 4th, and this one was written on November 2nd. So, actually, Anthropic might have even read this article that you're looking at right now. And oh, no, sorry. uh this article and actually took some inspiration from it. So anyway, I just wanted to follow on from previous vid last week's uh discussion and give you some further reading into why MCPs suck. This is part of a larger conversation on my show, Command AI, which we stream live every single week. We discuss the news and all things related to AI in the world of design and web. Catch us next week and join in the banter.