Why MCP Servers Are the Future for Businesses
10 min read
MCP servers are starting to to get more popular for services to offer. As I have started to use these tools in my everyday life and for a few work tasks I am convinced these will be the future of how apps/services are integrated with each-other.
MCP, the technology and protocol itself are in their infancy and have a lot of problems that need to be solved. You can only have so many tools on at a time, auth is difficult, its not easy to add to apps… etc. but the base approach will change things a lot.
A Story: Syncing Two Calendars with Claude
Recently I was talking with a friend and she was talking about how she spends like a half an hour every day syncing her 2 calendars, her google calendar and another one that gets filled from an app she uses for appointments. She had tried one or tow apps that said it would keep them in sync, but they were not great and still required a lot of time to do and weren’t cheap.
I suggested we try something real quick, in her Claude setup we setup 2 MCP servers for the services then wrote out a quick prompt to look at this week in both calendars and sync them over (and add a naming convention so you know when they get deleted). This took less then 20 min, and it worked. It was a simple use case but the AI tool was able to take a 30 min task and drop it to a 5 min task (that is mostly just waiting).
Previously I would have suggested maybe I help code up a tool for her to help do this for fun, but that always has the issue that it will break whenever the services update, or new edge cases will pop up. It also would have taken hours to develop, and if I wanted to make it “reusable” it probably would have taken days to add all the features you/others might ant.
But with the MCP server, if my friend wants it to ignore all all-day events or birthdays or something else weird, she can modify the prompt in a couple min. This does currently have the down side that she has to copy/paste the prompt in every morning, but it’s way better then it was before.
Integrations Required
Now, what happens when when we use this in business? for large companies you can hire an army of developers to link your various servers and perform tasks that require information from these 2 services. But for small companies, you are limited to services that already offer an integration with your other services, or use something like zapier and hope you can get what you need.
As you add more services, you add more cross-integrations needed. IRL you don’t need to connect every service to every other service, but if you only have 6 services you already have 30 integrations you need to build. This is why larger companies end up with service/event busses, or rely on double or triple entry; Small businesses end up having all kinds of double entry systems.
But what if you could just plug all your services into a central hub, and just ask it to do something? if you hook events into a system that tells an LLM the event happened and provide it with the MCP tools it needs to do any take you might put to it, it can just do the task?
Reducing the Effort
Traditionally you need to version your APIs, and do those versions for even small changes. Then you need to notify people using the old version that you have a new one, and that you will be getting rid of the old version, and someone needs to re-write your integrations.
With an MCP server, so long as you don’t change the general concept of what your application does, the LLM can probably just deal with the change. Give it a few tests and you’re good to upgrade your API. And if you need to add additional instructions directly to the LLM you an add those directly into the MCP instructions.
Adding Functionality
MCP based systems also allow for an easy way for people to ask the system to either explain or do things are not needed on a regular basis, this is the job of so many business analysts, they have to go into any number of systems, then throw them into excel sheet or into some report, and they will only ever have to do this once, or once a quarter.
If you can just provide all the tools into a single LLM agent, then let it run for an hour and you not have a report that would have taken days? and all it cost you was $10 in OpenAI credits.
What’s Missing
As of this writing, LLMs and the MCPs you connect to them are a little janky, you can get them to do simple tasks, but more complex ones are a little rough. And trying to just get MCP servers connected to your LLM is a pain in the ass.
Once the ease of getting these systems connected becomes easy, and business users sop being worried about the AI stealing their data (don’t you already store all your documents on google drive or microsoft drive?), this will be a huge shift in how businesses connect data together.
And from a technical side, I think developers dont understand how LLMs will be interacting with them. They need to be less “rigid” in their setup, too many MCPs require you look up the userid via one tool then connecting that with another tool… why not just make all your tools take in the email address or something more “human”? you’ll also want to accept the user ids, but also accept emails… pretend your tools will be manually entered by a human, and provide “answers” like you would on a web form “your entry needs to be in the format yyyy-mm-dd” instead of a basic error.
Conclusion?
I’m, not entirely sure how to end this, but once these tools mature I think this will make integrating many systems together that would be very difficult previously, and I also think this will make it much easier for small businesses to compete with larger companies with their software not being required to integrate directly with SAP or whatever.
MCPs or a very similar technology will start changing how we interact with all our services very soon!