Untitled - January 23, 2026

00:00:00 Speaker: Hello again and welcome to The Mostly Unstructured Podcast, uh, sponsored by Keymark. I am Clay Tuten, the CMO at Keymark, and I'm Ed McQuiston, the CEO of Keymark. Thanks for having me. Yeah, we're doing this again. Yes. Yeah. We didn't do it so bad last time that they kicked us out of the studio. I can't believe it. There's still time. There's still time. Easily. Totally, totally time to do that. Um, so today we're going to be talking about MCP. You know me? Yeah, with MCP, one MCP. You know me? Yeah, exactly. And IOT and IDP and. Wow, the rest of them. Wow. You've been working on working on this a little bit? Yeah. It's great. I like it. Yeah. You're. And you're the DJ. That's the problem. Uh, well, I pretend to be. I'm sure this will get cut out of it. God, I hope so. So we're going to talk about MCP. Yes. Um, which is the mostly coherent podcast. Right. Is that what we're. Well, no. Mostly unstructured podcast. Right? Yeah. No. But in all seriousness, MCP is, uh, a game changer, right? So just to I'm going to play the, um, layman in this situation, you can be the smart guy and I'll this will be this will be interesting. You did that. Well thank you. Um, so I'm going to I'm going to play the role I know how to play, which is, uh, kind of kind of low key, boring, dumb. Sounds good. Sounds good. So. But MCP is model context protocol. Correct? Right. Which is about as helpful as HTTP. Right. Right. It's very important to me. Yeah, yeah. Very important. Yeah. Yeah, it doesn't really say a lot about it, but, uh, about itself, but really just kind of at a level set for everybody listening. If you're not familiar with MCP. Um, it's a it's a communication for AI in its simple form. It's it's basically a glorified API in a way. Right. And similarly, if you go look up definitions on what it is, you'll probably get something around the lines of it's like a USB cord, right? It's a connector. Um, I was talking to somebody here the other day, and they like, I like to think of it as a Bluetooth. Like Bluetooth. Yeah, that's not a bad example either. Right? It's it's a way for AI to talk to all the other systems that, that it needs to communicate with. Um, really was brought about just not not too long ago by Anthropic who, um, most people are familiar with Claude, uh, the, the platform that they created, the LLM. So that's kind of it in a nutshell. It's able to help AI communicate. You wrote an article. It's on LinkedIn. Yeah. Um, is is MCP a big deal, right? Yeah, I think that's how. Yeah, that's how it's titled. Um, let's talk through some of that. Sure. What's what's the big deal about it, really? Well, first of all, I like your your Bluetooth analogy because, you know, at the end of the day, your iPhone, if you're an iPhone user or your Galaxy or what have, you shouldn't have to know that your headphone is a Bose headphone versus an Apple headphone versus a, you know, Sennheiser headphone or anything else. It should just be able to use a common protocol and connect your headphone regardless of who made it. And in a way, that's a great way to think about, uh, about MCP. What's so interesting about it, to your point about the timeline, is it's hard to describe in words how rapidly all the technology is evolving around the AI conversation. And I think that it's it's been become so ubiquitous in certainly technology news, but really in, in, in the, you know, public setting that there's already, I think, a bit of, um, you know, people being tired, you know, of hearing about AI. Yet the speed at which it's evolving is it really is difficult to describe. So, you know, as you know, we had a company meeting back in July, and one of the things I talked about was, you know, some of what we talked about on the last podcast, which is this idea of, you know, really harnessing this, this, this gold that is the content at rest in, you know, a content repository, um, you know, for our customers that that being OnBase. But, you know, it could be SharePoint repository, it could be box or OpenText or anyone else in those is is all this sort of gold in terms of data that has previously been unavailable? Right. When we think about the idea of, well, how, you know, if I could lift that data out. One - how might I do that? Two - how might I connect it to data from my other core systems? And prior to, you know, really the last couple of months, when we spoke in July, I talked about the idea of, you know, various, you know, uh, connected application types that might allow you to to do that. And seemed to me at the time that ultimately, because AI is, you know, it lives on data, right? Data. Data is its fuel. It seemed as though, well, the data lakes are going to become sort of the one ring to rule them all. And then that would make sense. Then I'm going to build my agents on, you know, probably within those data lake applications, it would sort of make sense. Those would become sort of the platforms. When I read about MCP, it changed everything about how I thought about connecting data from multiple sources. What MCP does, you know, and it's in its roots, is much like our Bluetooth example. If I want to query my Salesforce, you know, for data prior to this, I likely would have to know n number of Salesforce APIs to be able to go get sales data versus marketing data versus service data, depending on how I'm using it, and custom integrate those, custom integrate those. So I have to know the individual APIs. I've got to create custom, you know, calls. And look, it's a lot easier than it used to be. Rest APIs are much easier than it used to be, but that's one application. What if I also have Workday? What if I also have? And on and on and on. So now I'm just, you know, my life becomes building connectors to support, you know, some some end goal. What MCP does is it really takes the, um, the, the uniqueness of having to write to this API and this API and this API by creating a standard protocol. So all assuming that a vendor has an NCP layer, which vendors are sort of racing to do, because they certainly they want to have a seat at the table in the AI game. Now all of a sudden, for me to reach my end goal, I don't have to know every API under the sun. I can make these calls to these various, you know, Hyland. You know, they created NCP layer for OnBase, for example. So if I want to query into OnBase, I don't have to know every API of it to to go do that. So that really changed. I mean, to me it's a seismic event in this AI evolution because one of the openness, right. It's very open approach. The second is that now where I was thinking that the data lake becomes the stack, right? It becomes your agent builder, it becomes your analytics, it becomes your automation. It becomes because that's where everything is. I think that changes a little bit. And I think you start to look at the idea of those can be independent, agnostic tools that query your core platforms, and you can build your agents there. You can build automation, you build analytics there, and you can do it much, much more easily because MCP exists and MCP is kind of like the hub on the wheel. It's just what the spokes go into, all your different. What's interesting is recently I was, as you know, was participating in a, an AI collaborative. Yeah. Um, and when when MCP came up, the panelists just all lit up. It was really interesting just to see the energy behind it, because up to this point now we say that like up to this point, what, two years we've right really being into this thing, trying to make these connections. But uh, it is such a game changer and a lot and, and we've seen a lot of the other language models, other developers outside of Anthropic really jumping in on this, which is a sign of of how important it is. If you're going to have an you're going to add an MCP, right? Are are those done individually because you kind of mentioned that Hyland and creates sure that MCP for for OnBase or whatever. Uh would Salesforce create their own. Is there like a universal one? How should people think about that? Because I think they a little bit confusing. The vendors I think, are creating their own layers, and it's because that vendor understands its back end API and what they want to allow to kind of like A Rest API, like you're setting what it can and cannot connect with, right? Yeah. And you know, as you and the, you know, poor employees of Keymark know, I'm, you know, the king of terrible analogies and and, you know, it would be, you know, if, you know, let's think about use Salesforce as an example. Right. So, um, you know, Salesforce has many different clouds that they used to refer to them as. Right. So there's sales, there's sales cloud service cloud and marketing cloud. And they've evolved. You know how they talk about that. But if you sort of think in your mind, there's, you know, all these destination points. Each one's got a door in front of it. Each door had its own key. Right. What MCP is doing is saying, you don't need to have all these keys. I'll open the doors for you. Right. You just call me and tell me what you want, and I'll go open the appropriate door and get the information. And that's. You know, that's why really, the vendor has to create it because they understand what those keys are. But once they've created it, they've made it, you know, just so much easier for the outside world to consume. So it's really easy to say, hey, this is all rainbows and unicorns. There's some guy in IT going not over my dead body. You're going to connect in with this MCP right layer and have whatever willy nilly access you want. Sure. Um, that's an and that's an that's why we have IT people. Right. Of course to think through and and, um, leaders who are thinking through security and especially in I mean it's important to everybody but certain um, I think a healthcare, you know, and all of the pieces of so secure the information, you don't want anybody just being able to get in there. That's right. So having this open line to connect to anything kind of fraught with some risk as well. How is that being addressed? Yeah. I mean, you know, to your point, the you know, particularly the larger the organization, the more sprawl you have. We hear the term shadow IT, you know, quite a bit or, you know, I've got an application that is in my department that I manage and I'm like, they you know, they come in and show me this new AI capability and all you got to do is connect through MCP over here. All ought to be all great rainbows. But if somebody is freaking out because like, what do you do? Yeah. Without a question and and, you know, we at Keymark work in particularly highly regulated industries, right? I mean, most of our customers are in they're in government, their insurance, their finance, their their health care. And, you know, these are organizations where the security of their data is paramount to, you know, everything they promised their customers or patients or what have you. And so you can't just have this sort of open pipe, you know, so to speak, that says, oh, yeah, you know, query whatever you want. So, you know, to your point, there has to be sort of this governance component in that, you know, as to and this is a more dramatic term than I intend, but sort of policing, you know, the use of that so that we're not just, uh, able to share because, you know, in any one of those scenarios, I outlined the verticals, you've got personally identifiable information or potentially have banking information, you potentially have, you know, patient health information. And, um, the security of that is, you know, I mean, those organizations depend on. And so, you know, know, there's got to be sort of a rule set in place, access points. You know, those kinds of things. There's got to be, you know, a real, um, diligence, you know, from an IT perspective that this is not just this open pipe, you know, to go get data. Well, now I feel like I'd be remiss if we didn't at least talk about that. Uh, you know, being ISO certified ourselves. Yeah, we understand the importance of security. Um, and there's certainly people would say, man, you're just like you use a Bluetooth analogy is you just connect up to it, but you don't need it, you know, but it's not it's probably not that. That's probably in retrospect, maybe. Uh, no, it scares people. No, I mean, I think, you know, it's it's like, you know, I grew up as, you know, in the sales world. You grew up in the marketing world. I mean, we're not always the best at highlighting, you know, the, the, the risks sort of side of the equation. But, you know, it is, uh, and it should be a concern. I mean, it really has to be something that is well thought through in terms of who has access, what can they access? Um, you know, what are those rules and restrictions and all those things? I mean, you know, we work in a world in content where, you know, we we pride ourselves on the governance of what particular storing, you know, what we're storing, what we've called, you know, documents or content for for years is really just another form of data. And so who can access it now and all those things you've got to be very. So is that built into the actual MCP itself. Like is that so back to our references Salesforce. All right. Um, if they had their own they had MCP there like available. Right. They are they the ones that are setting that like where are the customers are going to set. Okay. Yeah. Yeah. The customer's going to own that. So it's not super complicated in concept, correct? No, I don't want to be the dead horse or just keep going talking in circles about the same thing. But what? What else are we? I mean, what else do people need to know about it? Other. I mean, because in a simple form, we've we've discussed that, I think. Sure. Well, I think, you know, there's some of the things that we've talked about before is that I, you know, I think our opinion certainly as an organization is the idea of I'm just going to go big bang AI, you know, kind of joke about, you know, I got to get me some of that AI. Yeah. Um, that's a, that that is not likely to lead to a lot of success. And, you know, picking sort of the right starting point with a very specific outcome in mind, you know, that kind of what is good look like. So if this is the end goal, I want, you know, I need to understand, first of all, what data do I need to reach that end goal? What am I going to have to feed into AI to allow me to reach that end goal? The understanding of that data, you know, what data do I need is then going to really allow me to understand what do I need to tap into, probably using MCP to make that happen. Yeah. And, you know, that's the thing that I think is. So I think it's why you saw that panel light up. Because all of a sudden, all those sort of locked doors that I referred to in the past in a way, are open. And, you know, if you're if you're a developer, right. That's that's the goal. And I and so I think the second thing people need to think about is that this really, you know, I think what it enables, and I don't want to say forces enables a customer to do is to think strategically about what do I want my sort of AI platform to be? When I say that, I'm differentiating between where is the data or data sources, but where do I want to build my agents? Where do I want to potentially build analytics? Where do I. And the answer is not going to be the same for every customer. Yeah. What I think this does specifically though is it creates choice. Because what it does is it enables the customer to not necessarily have to look at their platform vendor and go, I got to decide which one is the one ring to rule them all. You know, I'm either making my bet with, you know, my data lake vendor or my my CRM vendor or my. And I think, you know, what this does is say, no, actually you don't you don't have to do that. You can have sort of an independent AI stack that then taps into your, your cores and, you know, some vendors or excuse me, some customers are going to love that some customers, you know, really live by a principle of, you know, I'm platform centered and that's that's okay too. And if they make that choice, MCC still allows them that same flexibility. And that's what I think is super cool about it is you get away from this, you know kind of the past proprietary lock in that, you know I gotta make my choice and live with it. Well, what happens if you get unhappy with the platform now you've taken down your your platform and your AI. Right. Right Now you've got, you know, I think a lot more flexibility in how you approach it, which is sort of the magic to it. So back to our Bluetooth analogy. If your kid has an iPhone, you're on Android. You can still play your music through that one speaker. That's exactly right. And you're good to go. That's right. You upgrade and get smart and go to an iPhone, and I can send everybody who has a it's a great way to look at I mean really is you know, it is this idea of you now have, you know, kind of end to end flexibility. Right. You can choose whichever phone platform you want. Google. Great. You want, you know, Android, you want iPhone. And then what. Listening. You know what earbuds do you like. You know. And you've got all those choices and and I think there's it's a great way to think about it because that, you know, is what customers want. They want the flexibility to, you know, make those choices without being locked in. Yeah. Good. Well, I think that covers it. Yeah. No, I think something deeper than that. I would invite people, as I did last time, to reach out to us at keymarkinc.com if they want to start a conversation. We don't need to sell anything. We just want to have conversations like we're doing here. Unstructured conversations. Yeah. Yeah. I mean, I think, look, it would be, um, I mean, it's hard on us, the speed at which technology is moving. Yes. I, you know, I, I generally genuinely empathize with customers who are just, you know, seeing a cavalcade of vendors come in and saying, you know, my AI is the best AI. And I think part of our collective jobs here is to try to help the customer disseminate, you know, real from not right and come up with their strategy, you know, not somebody else's strategy that they're pushing to you. But and to do that, you know, it's what's my end goals. What you know, what data do I need to reach that end goal? How do I get that data, you know, and have a planful approach, preferably on something that you know is going to be a win for the organization. And once that happens, you know that whoever spearheaded that is going to become the most popular person, right? What can I do that with this? Can I do that with that? But do it based off that win? Not sort of a bet and a hope of, you know, we're, you know, to be able to say we're doing AI because that's not you know, that's not going to get you very far. Yeah. And it's moving so fast. Everybody is really learning as we go in a way. I mean, the idea that somebody is going to walk in and say, well, we've been doing this for. Yeah. Yeah. No. Yeah. That you have all the answers. I mean everybody needs to, you know, come to the table understanding that. Yeah, that's exactly it's a collaborative thing. It's a relational thing. And um, as we mentioned last time, pick a problem to solve. Don't start with the largest one. Start with something that is small and then gain momentum. We see that every day meeting happening. Came up in a meeting yesterday. Yeah. Where everybody's having success seems to be picking something that's doable and then learning from that. And then once that light bulb comes on, man, it's like a snowball. It just starts going and going and going. So because customers are really being forced to assimilate the idea of this, you know, kind of tech, right? It's just tech. Yeah. To the use case. Yeah. And that's hard for a lot of people. And you know but oftentimes once they see it up and running in their environment, even if it's different use case than, you know, their department, let's say to your point, that's the light bulb moment. Yeah. And I think finding that is really the key. Yeah. Well, thank you for joining us on The Mostly Unstructured podcast. Thanks, Ed. It's always, always fun to do this. Listen, if we said something brilliant, assume that we intended to. So we'll see you next time. Thanks so much.