Leap to Scale
Leap to Scale is for technology curious leaders of service firms who want higher margins without adding headcount. Each week, Justin Davis and Greg Ross-Munro show how to turn firm expertise into repeatable, sellable technology products: SaaS, packaged workflows, and AI-powered tools clients can buy again and again. With AI, more of your know-how can be captured, standardized, and protected as IP instead of being rebuilt in every engagement.
We focus on practical decisions: what to productize, how to price it, when to build vs. buy, and how to use AI responsibly without risking delivery quality or margin. Clear steps, real tradeoffs, usable examples.
Leap to Scale
What Is MCP (Model Context Protocol) and Why It Matters
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
In this episode of Leap to Scale, Justin Davis and Greg Ross-Munro break down Model Context Protocol (MCP) and explain why it is such a powerful shift in how AI interacts with real business systems. Starting with a simple explanation of APIs, they explore how MCP allows AI to securely connect to tools like CRMs, time tracking, and project management software so teams can ask questions, automate workflows, and take action using natural language. If you want a practical look at how AI can move beyond chat and become a true partner in scaling your business, this episode is a must-listen.
There we go. It's Friday.
SPEAKER_00Hey, it's Friday. Welcome in. This is Leap to Scale. It's the podcast where hey, we help service companies figure out how to grow past their headcount and uh and use products and product thinking to scale past the margin trap that service companies are very known for. And we know it well because we have worked in them for our entire career. My name is Justin Davis. I'm Greg Ross Monroe. Perfect. You jumped in there right in before I got my little transition sentence there. Um Friday. Uh Greg, how are you doing?
SPEAKER_01It's been a long week, hasn't it? It it was a long week on Monday, dude. Um, it is it is today is Friday. Um, I have a uh uh flight tomorrow morning, um, an event in Fort Lauderdale um at 3 o'clock tomorrow, a flight back to Tampa at 6 a.m. the next day, a car appointment at 10 a.m. This is just I don't know, this is not important, but we we can come back to the car here in a second. Um I've got to take it to the Tesla dealership. And uh after that, I have a um a Hanukkah festival event that I have to uh go to uh where my child is uh singing and dancing, and then after that, I have my parents' 50th wedding anniversary party. Um after that. So uh yeah, not much of a weekend ahead of me. So I'm gonna maybe this is a long episode, delaying the inevitable.
SPEAKER_00Hey, you know, that's all right. Yeah. Um it is a lot that you have. Um yeah, I don't I don't have near that kind of hacked weekend tomorrow.
SPEAKER_01You're gonna sit on, you're probably gonna sit on the couch and hack into your um golf radar system thing.
SPEAKER_00I there's a there's a chance that I do some of that. I'm also gonna continue my my quest against my garage. Um but since since um we both have laborious tasks tomorrow, let's talk about something fun today. Something fun and something new, something that we actually just passed the birthday outfit. We're recording this in early December of 2025, and um it just about 10 days ago, we passed the anniversary, the one-year anniversary of something called MCP. That is wild.
SPEAKER_01That is, I cannot it actually is crazy. Like I can't, I I find it, I find it so difficult to think about because in that year, um, people who are listening to us probably have no idea what we're talking about. No, but it's like going to know shortly. You and and yes, if you if you can continue to listen to the end, you will find out more value, blah blah blah. So yeah, listen. Uh but but the the point is like MCP is it's it's mostly the way the view to the world that we have right now in with systems and AI. Yep. Would you agree?
SPEAKER_00Yeah, yeah. No, I totally agree. Yeah, I think that's absolutely right.
SPEAKER_01Um so let's uh so it's so we don't have any scripts today, um, or any notes, to be honest. So um I will pretend to be the large language model and say, well, Justin, let's dive right in.
SPEAKER_00My favorite. Absolutely right. Yeah, that's a great idea.
SPEAKER_01That's a great idea. Why don't we dive right into this interesting topic of what is an MC, what is MCP and what is an MCP server? And seeing is can we let let's before we touch on that, can we just quickly, and this maybe is the wrong thing to do. Should we define what an API is first?
SPEAKER_00Yes, let's define an API because for some of you you may not know what we mean uh when we say API.
SPEAKER_01So I'll give that a shot and leave MCP to use. Okay, great, great, go for it. What is an API? Okay, well, let's dive right in, Justin. Um so uh computers or any system that is kind of computer run, you can speak to. It could be um your uh uh a machine that you have sitting, uh like a printer sitting on your desk or fax machine, heaven forbid. Uh it could be it could it could be a um another machine on the end of the the other side of the planet, which is a robot arm that you can move something around. Or it could be probably more likely uh something running on a website that you are like a mobile app and you need to send data to that mobile app and say, store this uses points for their game or whatever. Right? And the problem with these systems is they are written in other languages and they're they're they're secure little systems of their own. You don't want people to hack into them. So you can't just like send computer code to them and have it like execute commands. Yep. So what we have are called APIs, um which are uh what is it, access programming interfaces?
SPEAKER_00They are application interface. Application programming interface. Close, close.
SPEAKER_01It's something you use every day, and then um I don't know about MCPs. Okay. Uh so I wrote a book about this. Um so the so uh APIs are like when that one of those systems has a predefined way that you can talk to it without using programming language. And you can do that with like a text file. You could send the thing a text file, and the text file could have the format that says something like new connection, my password is whatever, and it gives you your password, and that password says, okay, you have access to do these things on my system. And let's say, for example, it's the it's a robot arm that uh draws with a crayon. And I want to move the, I'm gonna send, I send it a text file. Like I could, it's almost like emailing it. I email the thing, it's like, here's my access token and my password. I want you to move up to spaces, three left, three spaces to the right, down four, and then uh draw a circle. And it'll do like a little diddle thing, right? It'll it'll doodle. And that way you communicate is an API, and it's defined by the system you're trying to just talk to. There's a whole bunch of ways of doing them normally, like there's a whole bunch of standards around this, and we use them every single day. The whole internet is built on top of them. Every single mobile app that you've got in your pocket talks to some server on the internet using an API. Every, like if you have HubSpot or Salesforce or whatever, and you want to put something from your website or your mobile app, whatever, into your CRM, you can use their API, and it's published and there's documentation, and a software engineer can pick that up and they can follow the instructions and they can work out how your application can talk to some other application. How's that? Pretty good. Pretty good. Other than the fact that I didn't know what API stood for anymore.
SPEAKER_00But um yeah, like APIs are almost like the space in between the nerve cells. Like if you like zoomed out on the internet, everything is like talking, but it's like jumping over between things through these gaps, and those gaps are filled up with APIs.
SPEAKER_01Yeah, yeah. Yeah. So the reason why I I think this is an important thing to describe is because MCP is kind of like that. Kind of like that. Kind of not like that. Yeah. We okay, so can you dial it? And uses that. Yeah, it does use it. Yeah, exactly. Can you tell us what is an MCP server? Or no, what is MCP stock?
SPEAKER_00Yeah. So MCP uh stands for model context protocol. And you go to modelcontextprotocol.org if you want to learn more about the spec. This was developed by one of the leading um state-of-the-art model vendors. AI model vendors. AI model vendors, yeah. Um uh named Anthropic. And Anthropic decided. They make clawed, thanks.
SPEAKER_01Which I heard pronounced claw cloudy. Somebody said cloudy the other day, and I was like, what are you talking about?
SPEAKER_00That's insane. Yeah, don't say that. If don't say that, it's clawed. Um so so basically what they said is they said, you know, uh LLMs are great, AI models are great, things like ChatGPT. When we say LLM, we mean like Chat GPT. Um when um, but these AIs are great, but they're really limited to just the knowledge that they have in there. Then wouldn't it be really cool if they knew how to interact with the outside world? And they were able to do it in like a universal way so that any AI that you were using could figure out how to interact with the outside world. And that seemed like a pretty good idea. And so they came up with a as you do in the tech world, you come up with a specification, and basically you say, Hey, here's how we're gonna do it. Um and so they wrote up a specification for MCP a year ago, mind-blowing that it was a year ago, and said, here's what we're going to do. They integrated that into their own models so that their models understood it, and then that was adopted by the other model vendors, AI model vendors, uh, shortly thereafter. And so now it's a a pretty much a universal standard. Um, and it gets compared to the USB-C. People call it USB-C of uh for AIs, which I think is a bad comparison because that focuses on the compatibility and yes, it is a standard across all of the models, but like we have standards everywhere. We don't call REST APIs the USB C of databases online. I mean, like that's uh right. I mean, we could, but we don't, or servers or whatever. Um and and so I what what it what it steals from the focus that comparison is the fact that the powerful thing about MCP is you can plug it into another system with zero code, I guess depends on how you define code, but with a line of text, um, you don't have to know anything about code. Um and it is able to wire your AI up to your data in real time. Let me give you an example of how that works.
SPEAKER_01Give us an example, yeah, how of what this could what you could do with this.
SPEAKER_00Yeah, so uh I'll give you a practical example that um that we actually have used. We track our time in a thing called Harvest, and we have a lot of project data in a thing called Jira. These are two systems that a lot of people uh use in this particular type of service business, um, and web development services, software services. Um but you undoubtedly in your own world have your own like major behemoths, right? Um, and so what we what we needed was we needed a way to be able to understand like data between these two systems, or like be able to say, hey, what are the last 10 time entries that I have for this particular project? Or um could you take every task on this project and average out the amount of hours for it and do that over time? Right. And so these are like interesting things that help us to understand the business from a kind of analytics standpoint, but they're not things that we necessarily have a dashboard built for. And it requires us to bring data between two different systems and integrate it together. Um, or at the very least, go into one system and either hopefully find the report that they have or figure out like how to filter a custom interface and that kind of thing. So, what we did is we installed the MCP servers for both Jira and for Harvest. These are free to install. Way do you install them? You install them in you inst and the way that you install them, great catch, is um there's just a little file on your computer, and everybody has um examples of this. You can do it manually by changing this one file. And and um when you go on to Claude or ChatGPT and you look up how to modify the or install MCP servers, they'll have instructions on how to find that and how to change that. Some of the models now, like Claude, has the ability to do this via the UI um uh on the desktop application. ChatGPT does not actually, because ChatGPT still does not support MCP widely. You actually have to put ChatGPT in a special developer mode and agree to all kinds of caveats that they terrify you that like everything is going to fall over if you enable MCP. So that is an important footnote, actually. It should be more than a footnote, which is that while ChatGPT does officially support MCP in their own API that you can use programmatically if you're building software, in their consumer-facing software that you are logging into and using, there's still not really what we would call first class support for that, meaning it's not very easy for most people to use, it's hidden away.
SPEAKER_01Yeah, um, let me see if I if I understand this right, with the way you're saying it. Like if if I'm uh the way I interface with these things often is through like Claude or Ape OpenAI or ChatGPT or something that I'm like sitting in looking at, maybe it's through even Slack or Teams or something, right? Which talking to an LLM in between. And instead of like when I ask, if I asked the lot if I asked like open chat GPT, hey, can you um like tell me where where is a good place for um a client and I to have meet and have lunch, it would it would make a guess about that, right? It would have to guess. And it would it might ask you clarifying questions like what's your address?
SPEAKER_00It might say what days you have free.
SPEAKER_01Yeah. Where are you going from? Right. Yep. But but if that it had like an MCP server that it could reach out to and it knew how to do that and had the optionality to do that, it could say, let me just check your calendar quickly, let me check uh your hub spot to see where your client's offices are. Okay, cool. Now I'm gonna reach out to um the open table account, uh MCP server, and I'm gonna say, uh right, where are the five closest restaurants that have open table accounts? Now let me see if I can book one. And then it's gonna reach out maybe to your email account, and then it's gonna email a client and say, hey, I made a lunch appointment for us, or whatever it is. This time does it work for you? Yep. Yep. I can see why that's super dangerous. If the thing goes wild, then like it's got access to my email, and now it's booking account, it's like it's reaching out into the real world and doing stuff for me. That's that can be terrifying. And then in a non and in a non-programmatic, not like a like a non-procedural way, right? It's using it's it's using the intelligence of the model, but like we're giving the model little robotic arms that it can reach outside of its body for.
SPEAKER_00Yes. Right, right. And I think like those kind of tasks you do have to like you know, remember the AI is not just running on a loop, I mean, unless you set it up to do this, it's not just running in a loop all the time. All right, and so it is it is kind of bound, at least at the start, to the tasks that you prompt it with. Um, but it is good, it is good though, to have an eye on these things and to as you automate more and more, and this kind of a side discussion, but as you automate more and more with AI, and you get into what I call long chain automations. And I'm starting to kind of develop a theory around short chain versus long chain automation, and that AI really unlocked a thing called long chain automation. And when you get into that, you actually have to build in human in a loop points that actually naturally exist in short chain automation because the chains aren't long enough, and so a human has to step in. But in longer chain automations, you sort of have to work a human in a loop in. That's a little bit of a different discussion, but um it does relate to MCP because yeah, you're right. I mean, if you give it carte blanche access to your calendar and your email and you just say, hey, go into my email and camera and have fun, then then yeah, I would probably sit and watch it. Yeah. Pretty carefully.
SPEAKER_01But there's there's so many cool things that you could do with this, right? Like you could um you could you we always talk about intake forms, right? You could uh your intake could also then like it could go and do some research about the client that you know it could maybe look at what your pipeline is in um HubSpot and triage um, you know, where they should go in terms of uh how quickly they should we should respond to them. It can then maybe reach out to like your documents or your knowledge system or your CRM or something like that, your pricing or CPQ stuff, um, and be like, oh, okay, here's like a it could build like a first draft, and then when you wake up in the morning, in your Microsoft Teams inbox or your email, whatever, it's like, hey, this person contacted you. We think that you should do X, Y, and Z. I took the liberty of reaching out and working with the Liz pricing is and drafting this email to them. Do you want me to send it? Yep. Something like that. Yep. That would be pretty cool.
SPEAKER_00All doable. That is really easily doable, especially with like make.com. Right? Because now if you use something like make or you N8N or or Windmill or one of these automation canvases that allow you to kind of build these automations, these chains of automations. Um now that like with MCP, yeah, that stuff gets very, very powerful. Yeah. It's very useful both on demand and inside of automations and inside of software.
SPEAKER_01Yeah. And then what's cool, what's cool about them is we're seeing more and more people use them to talk to their own data. Right? Yep. Like the this these these things now go and connect to like their Salesforce instance and their um their ticketing system, and then um somebody who's not like a hardcore database access person or data scientist or anything can go and ask their system, hey, can you please like let me know which days what are like our most like heavy days um for our lawn care service uh in uh bookings in February normally? Because I want to know if I should staff up um with temporary workers uh at in what time. February's not a good time for lawns, I understand. But like you could now you could see. Okay, yeah, that's right. That's actually where I'm from. I should have known that. But the point is, like, you can you don't have to be a data scientist anymore. You can the LLM can reach out now through your MCPs, grab that stuff, and then do the analysis itself. Might even write code to do it.
SPEAKER_00Yeah. That's what I think like the real I think like MCP is very powerful for lots of things. And you can get very creative and you can and you can start to think about very sophisticated ways to use it to get really wild results out of it. But I do I do think that like the place to start with MCP is to go Google MCP server directory or mcp servers and or ask chat GPT or something to give you the top mcp servers for insert your industry. Um, and it'll give you that. And the chat with your data type of stuff, hook it up to QuickBooks, hook it, and no, it's not gonna go like it's not gonna feed your QuickBooks sensitive data into the model. Like that's not what it's going to do by doing that. Um, and so uh so hook it up to these things, hook it up to your time tracking, hook it up to your ticket software, hook it up to your invoicing software, hook it up to your CRN, hook it up to your CMS, whatever it is, and start to ask questions uh against it, and it gets super, super power, and it immediately will show you like the power of it, and then you can start to think, oh, well, what if I could then not only could I sort of get information back, but how might I now push information out or control things like the robot arm that you were talking about at the beginning of the episode, right? Like now I might say, Hey, I have this conversation from a meeting, and I want you to read it and create user stories and then stick them in Jira for me. And now you have a partner that is actually doing work for you, and you're just sort of controlling it from the the LLM, which is how I think the future of work is going to look like in general. Okay.
SPEAKER_01Well, I have a question for you then. So, like when when the i like we'll we all always used to use websites for everything, right? Yeah, that's what we were doing. Like, you want something, you log on to a website. Then the iPhone came out, um, everyone got smartphones, and then there was an app for that, right? App for everything. Do you think that in I don't know five years, every business will also have a little MCP thing that they you can install?
SPEAKER_00Oh wow. That's an interesting question.
SPEAKER_01Um MCP is a terrible name for it, by the way. Um if apps are cool, websites are cool.
SPEAKER_00A plugin or a a gem or a well gem is even not gonna bad, but yeah, I that's a really interesting question. Because I I do I think it'll be as ubiquitous as a website? No, probably not. I don't know.
SPEAKER_01I mean that's even what about your clients, right? If I'm a CPA or an attorney or something, maybe I can send you mice like, hey, this is this you can you can now access some of your data that lives on my system in some way or ask me questions through really, really interesting concept.
SPEAKER_00I that I have not thought about, and and now that makes a lot of sense because you know you could say, for example, your doctor in the future, or your attorney, or or something even less critical, right? Could say, Yeah, your app development company. Your what?
SPEAKER_01Something less critical, isn't it like your app development company?
SPEAKER_00Right, right, right. You could say, here's your client ID or your uh your user ID or whatever and password. They can give you authentication. The MCP is gonna handle the authentication for you. That's been a critique about MCP, but that's support for the first class support for that is going to be coming. And so um it'll handle that authentication, and then you will you will get your data done by going to their website, you will get their data by talking to it and getting some pre-made reports that get dropped in via MCP. I absolutely do see that world coming. Yep.
SPEAKER_01Yeah, I was thinking about this the other day. It's like, imagine if we could give our clients a little MCP thing, and it would say, Hey, how many hours have they um we spent on whatever project this month? Yeah, and it would it would be able to like inside ChatGPT or Claude or whatever environment they'll they want, they would be able to ask us questions that they would normally ask a product manager. It's so fascinating. There's some terrifying things about that, like how you'd have to gate that to make sure that you don't there are some nightmare people scenarios, sure.
SPEAKER_00But but that is a really fascinating I like usage of MCP and sort of it now that you say it seems very obvious. Um it's giving everybody a direct port straight into their company in a private and secure in the future. People who are listening to this and critiques of MCP are gonna laugh when they hear me say secure in MCP. No, right now there are some security things that have been widely talked about. A lot of those have been addressed through patterns, uh, as well as updates to the spec itself, and that will continue to happen. Security is going to get solved. It's too important, it's not a thing that will just be overlooked forever.
SPEAKER_01Um So should we make should should we should you and I like next week go try and build an MCP server for our clients?
SPEAKER_00Well, I think, yeah. I mean, I think I think we should. Yes, I think we should. Yeah. Okay.
SPEAKER_01We should we already we already have some MCP stuff for our internal people, and that's normally where I think most people are gonna start want to start. Yeah. It's probably a little, it's probably a little scary right now to give, like be able imagine, imagine if you are like I'm not like a lawn service company or your CPA or your attorney or whatever. And I can go, you can go into Chat GPT and say, hey, how my billable hours for this month, or when are you guys next coming? I know it's the middle of winter, so when are you coming to mow the lawn next? Because you take now every two weeks off, something like that, right? Some of those might be scary because you're worried that it's gonna leak other data. But your internal project people, your project managers, your um your I don't know, the people who are running the staffing for the day, um even like your lot your frontline staff, um, having access to tools like that where they don't have to go and you know go to like a manager and get that information from somewhere they can just go and ask an internal system and it will tell them that's very powerful and very cool.
SPEAKER_00Yeah, yeah. Absolutely, absolutely. And and the the craziest thing too is that like this stuff is is free. Um like an MCP server for data Jira's MCP server or QuickBooks or whatever does not cost you money. You literally it takes two minutes, and this like it is one of the most like insanely easy things to do, and the kind of like ease of installation versus value like benefit that you get ratio is I guess cost benefit ratio is just insane. Insane.
SPEAKER_01It can get it can take a lot of work to get to build an MCV service on like your side or building.
SPEAKER_00Yes, if it's own and installing it and creating it for customers, that's a different story. But for them to install it, yeah, it is cake.
SPEAKER_01Okay. Yeah. Well, how about that? How about next week? You and I we're we're you know, we got busy weekends. Maybe we should uh sit down and start hacking on something.
SPEAKER_00Yeah, well, sit down and write some prompts for cloud code to start hacking on.
SPEAKER_01Even better, yeah. Okay. All right, so if you if uh if you want to start um playing with MCP servers, go look at the MCP server um directory, just go Google Google that. Um go install a couple, like follow the instructions, or even better, don't follow the instructions, ask your LLM how to do it. Yeah, it'll tell you.
SPEAKER_00Yes.
SPEAKER_01Um hack around with a few, play around with a few. And then if you have an idea on how to like create a really crazy one or an internal one or something, you can give us a call. Yep. That's all you have to do.
SPEAKER_00Love it. Good stuff. All right, Greg. Well, uh have a relaxing Friday evening ahead of your insane Saturday. Thank you. I'm gonna go drink some wine with some some friends. So that sounds lovely. I think I'm gonna run to the liquor store myself.
SPEAKER_01Great idea. All right, my friends. Um I where why doesn't the liquor store have an MCP server that you can just go into chat if you didn't say it? Right ABC, I want you to bring me some wine, right?
SPEAKER_00I mean, really, my cameras in my house should be monitoring my wine and other shortage, and then should be reporting to me, hey, it looks like you're out of wine. It's Friday night. Want me to place an order for you?
SPEAKER_01Yeah. Do you do you want me to stop the order? Is probably what it should ask me.
SPEAKER_00Yeah, yeah, yeah. I was gonna say, I would set that one as dangerously skip permissions.
SPEAKER_01Uh my friend, we'll have a good have a good weekend.
SPEAKER_00All right, we'll see.
SPEAKER_01Bye.