I refuse to get on the hype train, and I regularly archive my rants on artificial intelligence before ever publishing them because I feel like I am screaming into the void. However, I am neck deep reading Timothy Snyder’s work right now and I feel regular dissent matters to counteract the negative consequences of technology in our world right now. With this in mind I will be posting regular posts to make sure my voice is heard in this moment, and today I will be sharing specifically why adopting MCP is such a bad idea. This round I won’t even pile on with the many other technical reasons why MCP is half baked, I am going to point of the business and politics reasons that every other single API pundit out there seems to take pride in ignoring.
- Database - There is a reason why we have been designing and evolving an abstraction layer over our backend databases for the last 20 years, and what on god’s green earth makes you think it is a good idea to just open your database up to the people wielding AI in this moment.
- Hype Train - Everyone is doing MCP. Everyone is talking about it. Your vendors all spent the weekend getting their MCP server up and running, with a press release by Monday—why do you think they do this? Is it to benefit you? No—it is get access to your data.
- Technical - MCP is a very technical specification that speaks to technical folks who are isolated within the technology echo chamber and is intentionally devoid of the business details we’ve been cultivating in the API space because getting access to your resources is the business imperative.
- Artificial Intelligence - You can honestly look at how AI is being wielded right now in government and the venture-backed technology sector and think, “hell yeah, that is what I am in support of and want done to my company!” MCP is a diamond-tipped drill bit designed to chew up your enterprise and consume all of your resources.
- Proprietary - MCP is open-source created by the “good guys” in a binary battle for power and control over the future of technology, there is no reason that you should be baking that into your operations, when you can do everything it does using OpenAPI and JSON Schema.
- Complexity - The promise of MCP is just another layer of complexity on top of the last 3-5 promises to fix your complexity that also became layers of complexity, and does nothing to help you make sense or obtain control over your digital resources.
- Costs - After everythign we just went through with the cloud starting off as being easier and cheaper and then our cloud bills going through the roof once they have us as a captive audience–why on earth do you think it is going to be any different with AI?
This is a hill I am very willing to die upon. If you look at out at all that API sprawl and chaos you have accumulated throughout this century and are in desperate need of an MCP fix to clean it all up—you should do it. Adopt MCP. Then move onto that next job once you realize what you’ve done, or I guess stay put and enjoy the defensive position you have established for yourself. However, if you are still interested in getting a handle on the digital resources and capabilities your enterprise depends on and provides to partners and customers, then I strongly urge you to look the other way in this moment. Any urgency you feel in this moment is intentionally manufactured to get you to expose your raw database and other digital resources you have already defined as APIs to the insatiable appetite of investors behind everything AI in 2025.
