Warning, I am not all for the hype on MCP technology. While MCPs are impressive and in some cases useful, I don’t believe it will be as useful as certain hypes make it to be. I think there is still a place for more traditional ways of doing things, because we know they work and they are increasingly becoming more solid.
I am a little biased against MCPs, because I am disillusioned to LLMs. If a technology is going to be useful, it needs to solve a distinct problem. There are several things I see wrong with the industry driving LLM technology, but this post isn’t about that. Rather, I want to talk about how useful and practical MCP is. Other technologies deserve just as much praise as MCPs get, but they are often overshadowed by MCP technologies because they don’t create as much of an awe.
Let me be the first to tell you, I am just a normal person. I do have opinions and biases just like anyone else. I also have my own life experience, and it permits me to see the world around me a certain way. The way I see things may be different from someone else, just like someone else’s opinion may differ from mine. This does not mean my way is the only right and true way, neither does another person’s opinion mean their’s is the only right and true way. What I mean to say is there is value to be found in collaborating among so many unique life experiences. This is what a community may look like — active, hard working, and caring communities.
Hype Technology
There are technologies that get hyped, and then the excitement slowly fizzles to a more sustainable level. New technologies go through their marketing cycle in order to encourage investors in the company. This is a problem, because I believe this makes money the deciding factor for what is made. Now true, making money helps sustain everyday life, but I am pointing to the level of money making for the sake of making money.
It seems a bit circular to make money for the sake of making money. Would I rather have explosive risky growth that can yield a lot of money, or would I rather have sustainable measured growth with something that probably will have lasting niche impact? I would prefer the measured sustainable niche growth, because I know that I may find fulfillment and sustainability so much easier there.
Some hype technologies are trying to sell their product without having a problem to solve. This manner of marketing just isn’t going to prove the product’s sustainability. If a seller is trying to get me to buy something, I would want to know how useful the thing is going to be. I would really like for something to not just be gathering dust for a long time, because the excitement of the thing will fade away.
In my opinion, here are some technologies that I think are a bit too over-hyped in their marketing. I do not mean to say that these things have no usefulness whatsoever, but rather their hypes muddle the actual usefulness. I mention these technologies to give context to my discussion on MCPs.
- Gimmicky Smartphone Features.
- Example: any phone with a zoom beyond 10x. I have trouble understanding why a normal user would want anything beyond 10x on a smartphone camera. At that point, just get an actual camera if one is concerned for telephotos.
- Graphene
- Was touted for its extra strength and high capacity battery.
- Blockchain.
- Despite its potential privacy usefulness for a decentralized system, it’s main reputation is just a gambling platform.
- NFTs
- A digital piece of art associated with blockchain technology. It rose in popularity, but quickly faded for how not valuable things turned out to me.
- The Metaverse
- A virtual reality that has all Facebook attributes? No thanks. The real world is just great to explore.
- Vision Pro
- While more or less practical, the cost of the technology is high for usability.
- Quantum Computers
- Awesome to get more computational power, but I’ve been hearing about this technology for a long time. Is it going to be commercial sometime?
- Foldable Phones.
- Cool technology, but it’s not very durable, it’s expensive, nor does it seem to be practical for an everyday phone.
Now, don’t get me wrong, some of these things can have legitimate usefulness. My issue comes from people using them for pure entertainment value, or these products are so costly to develop that no one would get it. I think of MCP technology has something similar to the previous products. There could be something useful about it, but the hype is seemingly drowning out the practicality.
MCP Technology
MCP technology seems to be gripping the attention of a lot of software engineers. It’s the Model Context Protocol (MCP), and it’s a technology to extend the capability and functionality of LLM chatbots with real data. It is impressive that one can extend a chatbot with something that seems like a REST API call, because that would mean a chatbot could get more accurate information instantly and do things on behalf of the user. This is a nice touch, I will admit. Even then I am holding onto a bit of skepticism for this technology. The setup and usefulness for MCP technology seems to be just as equal as a standard REST API.
I believe the hype for this new technology is blinding software engineers to whom this technology may really assist. Most of the hype that I’ve observed seems to be focused on making chatbots for a normal user. I see complex chatbots as an interface for a power-user, and I don’t see most users as power users. Another downside of MCP technology is that the tech is dependent on an LLM, and I believe that the limited usefulness of LLMs is becoming increasingly apparent. If one technology is dependent on another, then the dependent technology is limited and constrained by the other technology.
Not so User Friendly
There seems to be a vibe among software engineers that by programming a chatbot for the main feature of a website, one wouldn’t have to program an old fashioned user interface again. I think this viewpoint shows the limited knowledge of UX design and research of software engineers — me included seeing as my knowledge in that field is low. I anticipate that if a chatbot is the primary feature on a website, a common user will have to guess at what to do, because the features of website won’t be readily obvious to see. And as soon as a user starts guessing features then frustration grows. If there’s a frustrated user, then the user will begin to divest in the website.
A chatbot interface would have to cue in the user for its features, and immediately this seems to be going counter to the point of not programming a traditional interface. Part of the point of traditional interfaces is to provide enough cues so as to instruct the user how to accomplish their task easily. So if a chatbot needs to have “traditional features” why not just keep making a traditional website?
Or even better, conduct UX interviews and research to see where a chatbot could actually be useful, in order to narrow in on the usefulness of the chatbot. I can’t see a chatbot explicitly listing all the features for it to be useful. I’ve got a couple anecdotes for this at best. One, I consider myself a power user, and yet that doesn’t mean I take advantage of all the features that are available to me in the terminal and other developer tooling. I pretty much only take advantage of the features that are useful to me. By this, I can understand if I make features more usable, then that could be easier for a normal user. Two, Microsoft Word is a power user’s software, but for me it hits the same obstacles for usefulness. I only need about 10-20 features out of Word’s hundreds of features. A program like Google Docs is goated to me because of the reliability and essential feature set.
It’s odd to me to see a push for the experience of a website be oriented around a chatbot, instead of a chat being a complementary feature of the website. Having a chat agent be the main feature of website seems like it’ll lead to too much for a user to do. I think we are already seeing how too many features in an AI experience is a detrimental user experience. The AI-service companies that are winning are the ones that have targeted problems and focused solutions. Even then, as I have stated, I can easily see a chatbot being the main function of a website if and only if it was intended for power users.
LLM Usefulness and Constraints
I presented on LLM limitations at UtahJS Conference; essentially, while LLMs can be a productivity booster, they don’t boost all productivity. As things currently stand, LLM productivity is context dependent and non-deterministic. I’ve already seen too many demos where the demonstrator had to argue with the chatbot to get what it wanted. Aren’t demos supposed to be a medium so as to encourage the audience to use it? I’m not encouraged to use LLMs more by these demos — perhaps it encourages me to be more cautious to use LLMs, and for me, the caution is an expectation mismatch with the perceived hype.
Anthropic recently came out and said prompt engineering isn’t as useful as we think it is, and they suggested context engineering instead. This is just weird to me. Okay something we’ve been told to do is now not as efficient as it was originally supposed to be. But now we need to give an AI all the useful context in order for AI to do something useful. Once I read this, this sounded familiar. Isn’t this the same as duck programming, where a programmer talks the context out to a rubber duck? So Anthropic is essentially suggesting to do duck programming to an AI bot? At the point of giving sufficent context, couldn’t a good software engineer figure the problem out anyway?
I will admit, I could be fine with making LLMs more like duck programming. That way I am still engaging my problem solving skills, and then the LLM can do all the work. The only hesitancy I have with that is when the LLM actually does hallucinate, because LLMs are non-deterministic. There’s no knowing if I can reliably trust an LLM to not pull in something bad. Things would have to be done in smaller bursts, instead of doing sweeping code changes.
Conclusion
Now if MCPs were accessible without a chatbot interface, perhaps then this hype is worthwhile. I’m imagining a simple machine learning program, and not a chatbot, that can access an MCP server at any point. But then it enters the already existing competition of GraphQL and just plain normal REST. If I’m going to boil a chatbot interface into something more simple, like a button (😱), that already exists. Although perhaps an LLM + MCP can have an easy use case, but I believe that creating a website experience oriented around an MCP isn’t it.
There is one project that really calls my attention. Fabric seeks to augment AI UX into something usable. The description for the project puts the issue with LLMs home I think. “It’s all really exciting and powerful, but it’s not easy to integrate this functionality into our lives. In other words, AI doesn’t have a capabilities problem—it has an integration problem.” Only when AI is boiled down to simple and easy automation tasks, then it can really excel.