Part 2: Unlocking Advanced Insights with AI and MCP Servers

Transcript

Alright. We're gonna go ahead and get started, and I'm just gonna kinda go over our agenda and some some housekeeping notes. Today's section is sixty minutes long. We are going to email out this recording to everyone afterwards.

You'll also have a link to register for part three of our webinar also in that email. Please submit any questions that come up in the chat, and we're gonna get to them during the q and a. If we actually run out of time, we'll go in the community and post the questions and have them answered there.

So we'll make sure that you get an answer to your questions. I should have started by saying I'm Bethany. I am head of community here at Limble. I love all things customers and people, and so I'm super excited to have such a great lineup here and then have so many great customers and prospects on the call as well.

Our agenda today, we're gonna be going through introductions and the Limble community. We'll talk about MCP servers and Limble's investment into AI.

And then we'll see MPC servers in action with Yokohama tires with Kiri Rourke. And then we'll learn a little bit more about FM three sixty and their partnership with Limble, and then we'll have a q and a.

So as I mentioned, I'm Bethany. I am head of community here at Limble. If you haven't had a chance to take a look at the Limble community, I urge you to do that. It's a really great place to connect with our team and other customers.

Then we also have Gerald Wheaton on the call. He does consult he's a consulting engineer at f m three sixty. They are a long time partner of Limble, and they do a lot of really cool things with our customers. We have Kiri Roark.

He's the controls engineer, tire building at Yokohama Tire. And then we have Mike Scappa, head of AI and data strategy here at Limble.

This is my favorite slide. This is a little bit about the Limble community. What you'll find inside, there's peer advice. There's maintenance best practices.

You have access to the Limble team. There's product road map information and insights, and then we do a lot of discussions and live events. It is free to join. You don't actually have to be a Limble customer.

You could be just somebody interested in maintenance, interested in CMS.

It's a free online community for maintenance, reliability, and operations professionals. You can kinda see on the right hand side some of the information that's shared by customers and it's a super great place just to learn and be involved.

I'm gonna pass this over to Mike from our team. He's gonna talk a little bit more about, MCP servers, what they are, and Limble's investment into AI. Mike?

Great. Thanks, Bethany. If you can go forward.

So first, I wanted to talk a little bit about what our philosophy on AI is and how that manifests within the product.

So the first thing I wanna hit is is AI is an enabler. Right? It's not a replacement. Maintenance teams already have deep, hard earned expertise.

Our goal is to reduce the manual effort so that teams can focus on higher value work. Think of it as like a a force multiplier for technicians and maintenance managers. It's not an automation to replace people.

The second thing that we that we try to focus on when when developing AI functionality is really focused on practical AI over hype. Right? We try to prioritize real day to day use cases that deliver immediate value. Right? So think of it in terms of, like, if it doesn't improve workflow or decision making, we don't build it.

And the third, tenant of this is really, AI is kind of embedded within the workflows that teams are already using today. Right? So it shows up where it matters, whether you're managing work orders, assets, planning, analytics. We want AI to be a a part of the journey. Right? It's it's adoption should be natural because it's part of the existing user journey.

You may have seen some of these features or be using some of these features today, but but how does AI kinda manifest as product functionality and features?

So first is, again, right, trying to automate some of that that work. Right? Identifying and capturing data faster. So you may have used asset snap or part snap, automating planning. Right? So PM builder, being able to to incorporate AI generated PMs right into your maintenance schedules, anomaly detection, and then, of course, resource planning.

And then we'll jump into MCP a little bit. So not sure how much familiarity everybody on this webinar has with MCP, but but really think of it as turning Limble functionality into natural language. Right? So the easiest way I could describe it is, you know, you some everybody here is probably used to using ChatGPT or Gemini or Claude at this point.

But what the MCP server does is provide that bridge, that translation layer between Limble's API and the LLM, whether it's ChatGPT or Claude. And, so, basically, you're able to ask it things in natural language, find all critical open work orders.

And what that's gonna do is is determine how best to access the API for Limble, pull that information that you've asked, and then return it back to you in in natural language. So it's, it's very powerful, when used, and it's sort of it it it just makes the the the process of navigating work orders in in a natural language way a bit more efficient. So what I'll do is hand it over to Carrie now. He's actually built some integrations with the MCP, and he's gonna show us a bit of that.

Hi, everyone. Carrie Roark. I like they said, I'm controls engineer over at Yocom Attire.

You might be thinking, well, controls engineer, what does this guy have to do with anything to coding. Yeah. I just, put my I'm a person who sticks his head in places where it shouldn't be, and I like to tinker. I like to tinker. I like to learn, but I don't have a coding background. I have done a couple hello world tutorials in a few different languages. So I can look at code and not just be, have glossy eyes, but I'm not a coder by any means.

But through AI and agentic coding, you can pull off some pretty amazing things.

So Michael explained MCP at a technical level. I think it's probably good if I I was wondering if I need to show it again, but I think I probably should. I'm gonna show the same slide deck that I showed, last, webinar really quick, to get everyone on board with what a MCP server is.

Alright. Got me.

Alright. So screen yet.

You don't?

I don't.

I I have it up.

See it.

You do? Okay. Alright. So I'm gonna go through this really quickly this time. So first of all, you have an API.

That's an application programming interface. A lot of programs have these. It just is some it it provides endpoints and out points for getting and getting data and posting data to another program. So you have a bunch of little code snippets like this.

If you're using in a typical programming environment, you will take those endpoints, type them in your code, and that's how you get data in and out. Then, Anthropic came out with MCP. That stands for, model contacts protocol. So what that does is it takes all of this code that you have here and turns it into something like this, a straight up interface, a UI for the AI to use.

So instead of having to type in all these complicated commands and research all this stuff, the AI will see, get tasks from Limble as a single button that it has to push or put tasks into a report as a single button that it has to push. So at a high level, that's what an MCP server actually does.

So I started working on my own MCP server before I knew that Limble was building theirs. I just Limble has an a really good API. If you do wanna get into this stuff, they have a good API. It's documented well, so you can do things like build your own MCP server.

And that's what I did. So what you're seeing here, this is not this is Libre chat. This is not the MCP server itself. This is something that I use for testing the MCP server because it allows me to try a bunch of different models to see what works best with what I'm trying to do. So, anyway, I'm I'm gonna use a pre a canned sequence of events that I'm gonna type into the system just for time sake, but you can ask this MCP server anything you want. So first of all, I'm gonna ask it, oh, how many, work work requests did we receive last month?

And you'll see here in a minute, it's going to, it says ran, get current date. That's one of those buttons in the tool that I was talking about. It's running query task. It's grabbing that from directly from Limble servers. And here in a second, it's going to show me how many work orders work requests we received last month.

And it's important to know, this does take time. So it is performing a lot of tasks. It's grabbing a lot of data, and this is just a month's source of data. I can ask for things, like, a year in the past, but it's gonna take more time. So I'm gonna ask, how does that compare to the previous month?

So all of those questions that the bosses like to throw up in the board meetings that you were not prepared for, if you have something like this, you could just quickly ask it and get that data out in real time.

So here we go. Here's a comparison between the last two months. Here's February January versus February. Here's the change. This is really nice. I'm sure your boss will be happy to see that, but you can take it even further.

So I'm gonna say, you break these, down by department?

And this gets into something that I'm gonna talk about what my MCP server can do versus Limble's. So we have it broken down by department. And then, I'm gonna say, okay. Well, that's pretty nice, but, it'd be really nice if you can put that into a stacked bar chart segment and weekly by department.

So I can hit go, and it's going to do the same thing, run the tools that it needs. It's gonna visualize the data. And here in just a second, you will see stack bar chart with departments. I'm, over here.

Can you guys see that as the I don't think you guys can see the people there. But, anyway, you have the department keys. It's stack bar chart. It's interactive.

You can share this out, but, you know, I don't like that it's a light design. I want a dark mode chart. So I can come over here, and I'm gonna ask it. Well, that's great.

But can you change the design to dark mode color scheme and overlay a trend line showing the total work request?

The way that I designed this, you can ask for any type of color change. I wanna change the bars to this color. I wanna set the trend line to this color. I want four or five different trend lines.

I want it to be a pie chart. I want to have this font. If you can think it, you can ask it and and have it change that chart. So here's the dark mode design.

Looks like it missed the trend line, but I can go in there and ask for it to add it, anything that it didn't miss. But so this is one of the things that I added. The Limble standard MCP server does not have this visual data visualization. This is something that was really important to us and our company, and I was able to code it in.

I also something that it does you'll see that I was asking a lot of questions like, give give me a breakdown by department.

Now you might be thinking, well, Limble doesn't have a department. So how is he doing that? So my, MCP server has a custom context document that it looks at every time it runs a command. Or when you start up the MCP server, it seeds this into the MCP server, anything that you type in here.

So I have told it our in our organization, we use assets as department. These are high level assets. These are the these are the assets that rec represent department. So when I ask for department or reports by department, this is how you translate that.

Same thing for shifts. Limble doesn't really have, a way to break down reports by shift. We have a seven AM to seven PM shift and then seven PM to seven AM shift. So I was able to just tell that in our day shift to seven AM to seven PM, night shift to seven PM to seven AM.

Just like you're seeing here, there's no code. I just wrote it in natural language, and it feeds it in to the MCP server. So then I can just ask I need to know how many work requests did last night shift get or how many, work requests did last day shift get, and it can pull that data out. So that's the differences between, my MCP server and levels.

I'm gonna move on real Eric, I just wanna I just wanna hit on what you just showed there.

I mean, that is a super powerful thing, what you just showed. And, you know, I think if anybody's ever used LLMs like or Clawd in in advanced ways, you're familiar with managing context and and the power of it.

Not everybody sets up their Limble the same way, and and that's something that we know. There may be changes, like Carrie said, departments or different working hours.

Context gives you the ability to inform the MCP and and tell the MCP of how your business actually operates, and it's gonna use that information when it's when it's feeding information back to you and when it's querying. So definitely something to to click more into at some point, but but, yeah, it's super powerful.

Yeah.

Yeah. It's really enabled it to be used in great ways.

So, again, I'm not a coder, but I can come up with some really neat things. This is a dashboard that I designed for my company. It has a lot of our one, it allows us to look at the work orders by department really quickly. So if the department heads need to see what's going on in their department, they can look at the work orders here instead of in the Limble. They just click on whichever one they're in. They can do the same stuff, see the status of the of the work order, see what's going on there, print it out.

They can, set all these filters the way that they want and export that. I have a lot of the reports, that we need all the time here in widgets. So this is our weekly report. You can see work request, PMs, major breakdowns. You can, further break this down by department if you need to see that.

You can click on each of these widgets to bring up a chart, and all of the work orders that are related to that to that report are shown down here. Now if you need the report in a Excel document, you can click a single button to get an Excel document of that report. You can do the same thing with the PDF, a markdown. You can get a PowerPoint that shows, that, and I'll just show you that really quick.

One click, I get a PowerPoint that shows me that report ready to go, ready for me to edit further for our meetings.

And then in addition to that, I also added in, some project management. I was really proud of this feature right here. This is separate from Limble, but we were having an issue with managing projects. We had some other, that's all done through Excel documents and emails.

It was just really unorganized. So I made this to where, people can request projects. The when they request a project, it will go to the manager's screen, here, which will look like this, and he can approve the, the project. There's signatures that are required.

This automatically will send a notification to those people who need to sign this so they can sign off on it, collect the signatures all digitally.

They can track, what's going on with this project here, print it out into Excel document, PDF. They can change the, completeness of this. And when it's a hundred percent, it will send notifications to people who need to be notified.

Really I was really happy with that. And the same thing with some of our, forms that we need to get signatures on all the time with paper signatures. I'd integrated those in as forms that that the the signees get notifications when they need to sign something. So I was really proud of this.

So, if you I'm I'm gonna touch on that later. So, anyway, this is this project in a nutshell.

The last project I wanna show you really quick is, this new project I'm working on called Limble BuyBot.

So MCP is great, but it does take a little bit of know how to set it up. You have to find the settings in whatever, LLM you're trying to work with and just type in a couple simple lines of code directing it to an IP address. It's not very difficult, and Limble has really good documentation on their site, which I'm sure, will be sent to you guys after the webinar. But I wanted to make it even simpler.

So this is my new app that I'm making called Limble BuyBot. In this, all you do is open up the program, come here to settings, type in your client ID, type type in your client secret, and all your time zone. I have a space for that custom context, so you can just put it in right here, configure what type of AI you wanna use, your model, and you'll and what inference provider you wanna use. You set that all up here in a nice easy to use interface.

And then it's the same as just I showed you before. It's just you type to it. So this is that exact same sequence of questions that I asked the other one. You can see here it gave me the charts.

It put them here. And then, something that's neat with this is I'm gonna go ahead and ask one more question. Can you add that chart as a widget on my dashboard?

Click that.

See if it goes quick. If it doesn't, I'll just move on.

Carrie, while that's working out, I wanted to you know, this has been an evolution of how you guys have used the MCP and and, you know, it's been like a trial and error and, like, just kinda figuring it out as you as you've gone along about what works and growing on that. Is that correct?

That's correct. Yeah.

Because it can be overwhelming day one to see everything and what every everything you want. But sometimes you could just start with a small project as a foundation and kind of learn how it works and build from there. Is that kind of how you approached this?

Absolutely. And I get into that minute. The planning phase before you give anything to to an AI, that's how you get actually usable results. The better you plan, the better your results will be when you are coding with AI.

So this is done. I added it to a dashboard. You can see here I have all these different dashboards or widgets that you can add, dividers, and refresh this.

Think the back end just crashed, but I'm gonna move on. So, anyway, it's just a a automatically adds a dashboard, dashboard widgets there, and you can edit those further using AI, however you want. And I am gonna make it so that you can, at the end of the day, share that entire dashboard as a, stand alone program that you can so here's the chart there. It'll look just like this, and you can blow it up, export it out, download it, all that stuff.

So you could take this whole dashboard and send it off as its own little program, load that up on a screen that will pull this data live as you need it however you wanna lay this out. So this is what I'm working in now. It's kinda rough around the edges. This is an ongoing project right now, but something I'm really excited about.

Alright. So at this point, you're probably asking, alright. I keep saying coding with AI. What does that mean?

Well, agentic coding is the the technical term for that, and that's, just using AI to code.

So in the past, first, we had binary.

We had ones and zeros. And you would take those ones and zeros and put that into a translator or something or a compiler. And when you put it into the compiler, it would run all this stuff and spit out a program.

Then it got a little bit better, a little bit easier to read. You had assembly code, and it's a little bit easier to read. The average person can type in these things and know what's going to happen. But you take that and you put it into a compiler or translator, whatever you wanna call it, and then it would spit out the program.

Then it evolved to modern coding languages. These are a lot easier to read, a lot easier to understand. You have Python. You have JavaScript, and you can actually type in words now. And it's it's a lot easier to teach people, a lot easier to read. You take all of that code. You put it into your compiler or translator, and that is what spits out your program.

Now we have AI. Now you code with natural language, or this is the future. There's still a lot of bugs, so don't bite my head off. There's a lot, I understand there's a lot, left for AI to develop before it really completely takes over.

But this is where it's going. You have natural language. You can just say, make me a program that sends out a KPI report every morning at seven AM, and now your compiler is AI. So the AI will take that data, compile it together into program, and put it into a program.

And that is basically what AI coding is. You have a couple tools to do that with. The there's two big ones that I'm gonna point out here if you did wanna get started with this. The first one is cursor.

This is a this is a really great one to to start with. It's where a lot of people who try to make programs, start with.

It's it's good.

It has a very clean interface.

But, honestly, these days, it's hard for me to recommend that even starting out because the next one clogged code is just quickly becoming the industry standard. And there's it's there's so many powerful features that you just can't get anywhere else using ClogCode, and the models that Anthropic put, puts out are tailored towards coding. So you just get better results hands down.

So if if you did wanna get into this, I really suggest going with clog ClogCode.

So, I get how am I looking on time?

You're doing well. Okay. Yep.

So let me, I'm just gonna show you guys this really quick.

I'm going to just start up a terminal here. You can use ClogCode in a in a pretty interface, a UI, but I I like using it in the terminal most of the time. So this is what it looks like in the terminal. I have this little demo app prompt that I came up with just to show you something, that can be finished quick.

So I'm saying just build me a simple, Python app, a Square interface, two buttons. You click good. The the background will light up green. If you click no good, the background, will, light up red.

So this is how it looks when you're actually making a app. But I'm gonna paste that in there.

And this will take, in my in my testing probably around three or so minutes to finish. I'm gonna let that run, and we'll come back and check on our actually, it's gonna open up as soon as we're done because that's what I told it to do. So in the meantime, let's go over some tips if you're doing this. So first of all is design your UI in something that you're comfortable with.

If you don't have coding skills or you're not too technical, but you have creative skills, if you can use PowerPoint, if you can use Paint, if you can use Photoshop, and especially if you can use Illustrator, that can be very popular, very powerful. You can make things like this in PowerPoint and give it a screenshot, and the AI will be able to turn that into a real app. And it doesn't have to look pretty. The AI will pretty it up for you.

But if you can use something like Illustrator, in a couple weeks, you can take that, Illustrator skill and learn this program called Figma. So here you can see this dashboard here. I manually designed this entire user interface before I hit a single line of code. So you can see here's this, signature, area.

This is how I wanted the completed request to look when they're condensed. This is how I wanted to look when they expanded. You can see the components. I even made the, the percentage, clicker all from scratch.

So you can do all of that from scratch. And when you do that, I can click on this, and Figma has an MCP server that I can connect to my agentic coder, and it will pull in that code directly from Figma and recreate this very close on the first go. So if you do have those skills, leverage them. You'll they'll really help you out.

Let me check on our code here.

Okay. It set it. It it finished up. Okay. Here we go. So here's our program. Our good good, not good.

I click good, makes the the square green. Click not good, makes it red. So that's all you gotta do is just prompt it what you want to make, type it in, let it go, see what it does, and then iterate on it. That's the the basic workflow.

Couple tips here. If you do get into this, treat writing your prompts as if you're writing code. So I am when I'm making it, I am meticulous. I say, when I click this button, I expect this to happen on this page on when I see these tabs on the top.

When you click this tab, it's going to be this page with these buttons. Click this button. It does this. Be as detailed as you can, and don't just type it right into AI.

Type it into Notepad or or chat back and forth with a separate AI about how to structure this. And after you have it all in line with how you want things to go, then you take that entire prompt and put it into AI. You're gonna have a lot better results. There are as you get in deeper into this, there's some there's some really nice workflows that people have come up to expand on this.

But in a nutshell, if you do that, if you're just very detailed at the start, you're gonna have better results than somebody who isn't.

Next, ask AI to include inline comments and manually review the code that was written yourself. Don't just sit there, go off, have cop, coffee, and just let it do its thing. You can do that. It's gonna put out something, but, that's when you'll end up with the AI slop that, you click a button and it doesn't do anything like what you wanted, or you click a button and it deletes a database that you were not expecting.

So this is really important. Watch it. See what it's doing. You will start to pick up some coding techniques too as you do that.

Something that I didn't put in here that's really nice, a nice AI trick is if you're wanting it to explain a really technical something really technical, ask it to explain it to you like a smart fifth grader. If you don't do that, sometimes it will give really childish analogies. Like, if you're like, I need you to break this down simple for me. It'll be like, okay.

Well, you know, a hard drive is like a refrigerator, and your files are like mayonnaise and just just just silly things like that. So telling it to explain it to me like a smart fifth grader will make it so that it's easy to understand, but also just not just too simple.

So, anyway, I think that is it. That's everything that I have.

Thank you so much, Carrie. I wanna ask you. If people took away one thing where they could start today, what would you challenge people as as one thing? If they've never done any of this and they're wanting to get started, what would you suggest maybe they they start with? What do they try first?

Don't start with something big. Do something that you understand completely. Try to mimic a app that's already out there. So let's say something simple like a reminders app. Try try your hand at doing that. See what it's like to build a reminders app and, just go from there.

And then after you built your first reminders app and you really feel like, you can dig your teeth into something substantial, just do it and just understand that you might have to, a few times, delete the entire project and start all over. That's just part of the process, and you'll learn workflows as you go.

Awesome. I love hearing about everything you've built and that you're self taught and that, you know, you were curious, and this is kind of how it's manifested itself.

So thank you for sharing everything that you're doing. Mike, I wanted to Mike Scapp, I wanna touch on one thing. The work that people are doing with the MCP does not impact the data that is in Limble or their their their actual setup in Limble. So if they make a mistake, you know, trying one of these agents out and building anything, they're not going to impact their setup in Limble and things like that.

Correct if you're using the official Limble MCP server.

Ours is is locked down, so it really only is aware of read access.

But but what Carrie and Gerald did, you know, they they kind of rolled their own MCP server, which has a bit more access. Now look. LLMs are inherently nondeterministic. So, you know, future results, aren't guaranteed.

Right? So I think it's, you just wanna be careful if you roll your own. Right? It it might be smarter to to stay much more of a of a whitelist approach.

Thank you.

Let me add that that one more statement. If you are doing like he said, if it's something very critical, AIs make mistakes. As much as I am very proud of myself, what I've been able to do, AI coding is still in its infancy. So if you do make something that's dealing with mission critical data, it's very important you can use it and make make your program, but have somebody, like John or Gerald, f m three sixty, just get in contact with with them and have them look over what you've created. They might be able they can turn into something that's secure, safe, polished. So something to keep in mind.

Awesome. And that's a great segue. So, again, Carrie was self taught. He he's child in Arab.

He he was curious and kinda built this out on his own with the support of his company, and has engaged other stakeholders at his company. But another option that we have a lot of Limble customers using is our partnership with FM three sixty. And we have John and Gerald here from FM three sixty to kinda talk about the partnership and kind of how they, see people using AI. John or Gerald, do y'all wanna take it away?

Yeah. No. And I wanted to say reiterate, like, what you said, Bethany, of what Carrie's done with being self taught. And that's one of the things I love about facilities, and I always joke that, you know, we're we're very creative. We're like the modern day MacGyvers within facilities. And, if you're if you're in facility management long enough, you get two things, a thick head and thick skin. And part of that thick head is, you know, Carrie probably had to beat his head against the wall quite a few times to figure out how to do this.

Nice thing is, technology is advancing very, very quickly. And, you know, you've got really, really smart guys. Like, you know, we've got Gerald on my team that we is a recent addition, that is, you know, continuing to open the door of how do we work smarter to leverage technology so that we can be more efficient in how we use our resources, how we do our jobs, how we sell the value of what your department provides to your respective organizations. I mean, that's imperative.

It's like we have to. And and we as FM three sixty and the partnership that we have with Limble is, you know, we come in, you know, providing the boots on the ground, providing some of the the the brains behind the keyboard, but doing this overhaul overall facility management and operations and maintenance consulting services. So, you know, example was mentioned earlier, the, taking advantage of the asset snap and inventory snap. Our team's been using that in the field to basically you know, as we go in QR code, add assets in the field, taking advantage of that snapshot of the nameplate and letting, the AI as part of, Limble's mobile platform extract that nameplate information, oh my goodness.

It's huge. And then, of course, the AIPM generator, being able to take advantage of that functionality, which my team has also been using for a number of our different clients and have had good success. So these tools are readily available for you. Start taking advantage of it.

And then, you know, the other thing I wanna reiterate before I open it up to Gerald is the the data you're trying to get out using MCP and using AI is only gonna be as good as the data you put in.

So I cannot emphasize that enough of making sure even just something as simple as location and assets. You know, your nomenclature, do you have all your asset information in there? Are you capturing all your work orders?

Mean, more than half of the facility organizations are more than fifty percent reactive, and many of them don't capture corrective work orders. So if I'm now asking AI, hey. Please go tell me how we're doing, it's only gonna be able to give me half the picture.

So making sure that you and your team have a good solid foundation of data to start, you're fully leveraging the robust tools that that Limble has of being able to capture this data in the field, creating work orders on the fly, leveraging the QR codes so that now I've got the data that you can start using to drive efficiency within your own department and then, of course, selling the value of what your organization of what you provide to the organization.

Gerald, anything you wanna add into you know? I mean, I know you've been dabbling around and kind of opening up my mind to some of the cool things that, that Limble and, MTP can do.

Yeah. Yeah. Absolutely. Well, I appreciate that. Well, you know, I'd I'm not entirely sure what everyone's sort of relationship is with AI in terms of what you've used it for, how much you understand about it. Carrie had some awesome things to say about Cloud Code. As a software engineer myself, I've used that extensively and very familiar with the wonders that it it it can do as as Carrie kind of laid out there.

And if I could speak to a different side, maybe as as far as what what Limble's MCP now means for, people who maybe, as as Carrie said, can't code or maybe for for those of you who just couldn't care less or don't have the time to or, like, even if you wanted to, just not really relevant to your to your job. Right? What what AI is doing, I think, in general and what the what Limble's MCP is doing now for your data or can do is it's sort of forcing a paradigm shift, right, to where we now have the opportunity to sort of rebuild a lot of these SaaS products that that we're renting from other people ourselves.

Right? Or we can now transition, and you can do do both of them. But one of the paradigms that it allows us to sort of move into is just go straight to what it is that you that you want. Right?

Instead of paying for a company to build a dashboard for you and and do all this, just go ahead and ask the AI for the information that you're gonna use the dashboard to in in inform you about. Right? The whole point of gathering information and putting in in these dashboards is so you can make a call about your data, so that you can be well informed. What this MCP allows you to now do is have an assistant that's conversant with your data in your pocket, potentially, and I'll speak to that in a minute, or on your desktop.

Right? Anywhere you want. I have a couple different examples here. You know, from the very top level as far as, like, you know, a a facility's director and then and then manager kind of getting information about how are our our work orders going, what's the asset cost, you know, where are we losing money, how are things how are our technicians doing with with knocking off PMs and all that.

You can also apply this at at a more kind of boots on the ground level for your actual tech technicians. Maybe one guy goes out to fix some asset to do his PM work. And while he's out there, he could whip out his phone and just throw a quick question into, say, Claude, or ChatGPT, whatever your LM of of choice is.

Granted, that that that you've connected or provided that you've you've connected this MCP. He or she could could go ahead and ask, hey. Are there any other assets in this area with outstanding, you know, work orders that I could kind of attack while I'm right here in this building, in this section, whatever it is. Right?

That's that's that's an example. That's an example of in the field sort of action that you can use with this MCP. Right? Now as as it was explained earlier, I think Carrie did did a great job of just of explaining API.

Right? API is kinda is deterministic access to data. And that's a very powerful tool when you have a technology team and then when that technology team is accessible all the time and to build any kind of workflow or dashboard you need. What the MCP then allows you to do is is to kind of get right to the conversation that you wanna have about that data without having to wait a long time to sort of build out all of these workflows and dashboards and everything to to sort of inform what it is that you wanna do.

I could go on and on about this. I'm having I'm I'm happy to answer questions maybe over email afterwards. We could connect over LinkedIn and chat about this. But, this is something that us here at FN three sixty, would be happy to partner with you.

We we would be happy to help you guys connect your MCPs and kind of learn about how to use this, what kinds of questions can we ask, how can we create sort of, thoughtful workflows that aren't just giving us data, but that are maybe emailing us when when a variety of different things happen that we don't have to maybe pre tell it to to be aware of. So that's that's kind of a lot there. I'm happy to answer any any other questions. But generally speaking, I think this is awesome.

Limble's done a really incredible job with their MCP at this moment, and I'm sure that's only gonna get better. So I'll I'll be excited to keep using it and then helping you guys get into it as well.

Awesome, Gerald. Thank you so much. So we're opening up the chat for any questions that you have on MCP AI, really anything Limble. One of the things that I think, lots of that was touched on, a little bit in this, meeting was that Limble approached AI as not a replacement for your your team members, your technicians.

And I really think when I'm out there on the ground talking to maintenance professionals in in real life, that is a real concern that people have is that we're building things that are gonna replace them. And and so one of the things that I really value about Limble is that we have met with customers as we're building our AI, and this has been, like, from day one, our approach from AI has never been about replacement. It has been about, like, supporting the technicians on the ground. And I think that I just wanna reiterate that as as far as how we're approaching AI today and tomorrow in the future.

So super excited to see what else is to come.

Can I can I jump on that real quick, Bethany? Because and you're one hundred percent correct. Is that AI, you know, taking advantage of the technology and the tools is not something that's going to replace. It's going to augment.

And and, really, within the facilities and maintenance community, we don't have a choice but to embrace this technology. We know that we're having a very difficult time finding skilled labor, right, finding our wrench turners. And we still have roughly twenty five percent of our skilled labor workforce to retire by the end of this decade. That's scary.

And we've been wrestling with this problem, and we're not are we getting ready to round the corner? No. It's still going downhill. So we don't have a choice but to take advantage of this technology.

So, to me, it's it's not a problem. It's an opportunity, and we need to make sure we we put our hands on it.

Thanks, John. I agree a hundred percent.

I'm gonna check the chat. Heila asked in the chat, and this is more for all of the participants. Has anybody are y'all dabbling in AI? Are there anything with AI that you wish you could build for your team?

And while people are answering that, I'm gonna ask Robert's question to our panelists.

Will you ever have a native AI inside of Limble, or will you maintain only the m MCP interfacing to an existing AI? Mike, I guess that would go to you, Mike Stappa.

Yeah. I'm assuming by that question, you mean more of, like, a a copilot style LLM interface with inside Limble. I'm gonna reserve that actually for Laura McCarthy, our VP of product. I don't believe she's on this call right now, but she'll be talking about product road map and and what's coming in a in a future series. But I don't wanna I don't wanna get too much away here.

If I can speak to that, not from a product point, obviously, I don't I don't work for Limble. But one of the things that I've heard is concern about data privacy. Right? And, hey.

If I plug this MCP into Claude or to ChatGPT or or or Gemini, it sort of gets sent off into this magic black box that does wonderful things and sends us back kind of the product of these spells. Right? And not that you necessarily need to be worried about those those things, but as those concerns come up and you are uncertain or worried about privacy, that's that's completely fine. Setting up locally hosted LLMs or models are things that our team can also help you with.

At the same time, that is something that if and when Limble puts out something like that, would be extraordinarily beneficial and another way for you to then trust kind of further what Limble is is doing for you where you don't have to outsource that kind of data analysis to to any other company.

Thank you, Gerald.

Barry asked, can Limble give a demo on how this is working now that we can access?

Mike, are there any demos out now on the MCP? Are those things that are in the the works, how people are using them?

That's a good question.

I don't know the answer to that, Bethany. Well, I'll have to follow-up and see what demos are available if there are any created.

Yeah. And I know that in the community, Barry, some people are sharing kind of what they're building. It's not necessarily produced by Limble, but it's good to see how other customers are using the m p MCP. And I can also get with our product marketing team and see what we have available.

Barry's also asking a follow-up to, I think, Gerald, what you said, and it says privacy. Does that equal HPPA requirements? Was thinking HIPAA.

Yeah. The HIPAA requirements.

I'm guessing that's what Barry's getting That's what I was thinking as well.

Yeah.

Yeah.

So and this is often a concern, of course, working within the health care industry of, oh, what about HIPAA? But in my experience in working with a number of health care clients, we're not capturing any asset information inside of the CMMS. So it really should be a moot point. But as Gerald indicated earlier, with the MCP, you basically, you're using just your data. And if you're bringing in that that AI, the LLM internally, that's only using your data and you're not taking it outside of your database. So that's gonna be a key component there.

But now you if you can you shouldn't be keeping, hopefully, any patient information in there any inside the system, but even if you were, you should be able to limit it to your database.

Thanks, John. Mike, I I have another question for you. It's gonna be from Tim. For queried proprietary data, when the MCP server pings it, how deep is the query and how secure is the data that may be exposed from the query? Should we limit enter entering certain information to Limble assets trees?

Look. Anything that sits on the server on the Limble application itself at rest in your database is is covered by our standard securities. When you pull that information out with the MCP and it goes into to some LLM that you're using, and that's what Jeribel is getting at, that is governed by a different set of security policies.

Our MCP, the official Limble MCP is is part of you know, is covered under our security policies. But, again, you might be plugging Claude into that MCP. So Claude is gonna see that information that comes out of it, or OpenAI. You know, I think John and and Gerald were both hitting on that there's ways to create more private models for LLMs, whether that's running it locally or if you use AWS, your organization may already have a, you know, as kind of a an isolated instance of, of models running that that could be used. But, you know, assume that if you're pulling information out of it and you're on just a general consumer plan of of ChatGPT or of Claude, that it could see that information.

Thanks, Mike. And so a good takeaway here would be that it's probably best before you really start inputting data that that the the LMMSs are gonna have access to to speak with your legal team. Everybody sets up things differently as far as permissions and how they want data to be used and and how they want software programs to communicate with each other. So it might be good to just kinda bring in legal team, your own legal team to to kinda discuss how you guys approach AI. Curious that what you kinda did at first is is determine how the company was gonna approach AI and and how it would plug into the work you're doing?

I asked some questions, but, really, I was just keeping things just safe on on my side, just not putting any data that I knew was sensitive to the AI and always using reputable AIs. There's a lot of different AI companies. You have Anthropic. You have DeepSeek.

You have, the Alibaba. You have all of these others. And every single one of those AIs has their own privacy and data retention policy. Look at those that have, zero, I'd recommend that if you are gonna use a cloud based on them, look for ones that have a zero data retention policy.

And, also, yes, talk to your IT department. Talk to them. See what is allowed. See what can be done.

If you do need to play with this stuff ahead of time, I just put in a couple things in in the chat. If you wanna play with some local AIs where everything is a hundred percent on your local hardware, you can you can play with, LM Studio and Olama. Just check those out. That's two of the the big interfaces for doing, local AI.

And those, can hook directly into the MCP server. And then there is nothing else in the loop except Limble servers and the AI running right there on your own hardware. So there's a couple things.

Yeah.

I'll address that one from Robert.

So, Robert, yes. The answer is yes, but not as directly as I think you would think. Right? So, this is where Carrie was showing some context. And and and what you would probably do is set up context that says if the, you know, user asks a question about an asset in terms of, like, technical capability of the asset or troubleshooting the asset, things like that, the MCP would have context to first fetch the file from Limble and then explore the file directly in the LLM that you're using. So at that point, it's just, you know, ChatGPT or Claude reading a PDF. It just happened to be that the source of the PDF came from Limble.

Awesome. Thanks, Mike, and thanks, Carrie, for sharing those resources. I, appreciate everybody coming today. I wanted to give everyone who's on here a little preview of part three, the final part of this series. It's super exciting.

The next one is on April fourteenth, and it is talking about the ultimate guide to facility mapping and asset clarity. Deebo, John, are you available to kind of give a sneak peek into that?

Yeah. Sure. That sounds great. Okay. So for that one, we're gonna talk a lot about mapping, and I kinda have three things that we're gonna cover. The first one being what mapping options you can use in Limble, and I'll talk about a few examples of how I did that with mapping out street lights and various assets to help our new employees get around our college campus quicker. And then the second example actually, if you go one more slide or two more slides down, Bethany, I think we added it maybe.

Let me go back out. Oh, there it is.

Let me Okay.

Yep.

So do you So this is Okay.

Quick teaser. This one was built in Power BI, and so this is something I built for the college to be able to see a variety of work orders on a map. And so not that this is this is a static slide, but you'll be able to click on one of those buildings and actually see. And if you go to that next slide, it'll show you what you see when you click on one of them, but it will show you a report of what is in each building. So that's a bit of a teaser. And then I'll show one other strategy using Canva to build a request map that you could have requesters submit by clicking on an asset rather than going through an asset hierarchy. So that's a teaser of what we're gonna do for that next webinar.

Amazing. And we will be sending out a link with the recording, which will also have a link to sign up for the final part of this series about mapping.

In the meantime, we would love for you to come into the community.

All of our panelists here today are in the community. If you have additional questions, if you wanna see how other customers are building out things with the MCP or other parts of Limble, custom dashboards, all kinds of fun things, widgets, go ahead and join the community. Join the conversation. Kinda see what other people are doing.

As always, we are so appreciative of your time and that you're a customer of Limble. We wouldn't be here without you. So thank you for attending, and we hope you got something out of today's webinar.

‍