Optimizing Your Service Desk with Claude: Syncro MCP Workshop

Welcome and Introductions

Richard Dean: Hello, welcome everybody. Thanks for joining us today. We’re going to give everyone a couple minutes to continue to join, so just sit back, relax, and we’ll get started very soon. We’re going to talk about optimizing your service desk with Claude and Syncro today. So we appreciate your time, we appreciate the effort coming out to listen to myself and Jasper.

Richard Dean: All right, let’s go ahead and get started. Again, thanks for everybody for joining us today. I’m Richard Dean, Senior Director of Product Management here at Syncro, and I have been luckily involved in the development and the deployment of our MCP server, and we have worked closely with Jasper and ROI Technology to develop our first plugin. I am happy and honored to represent everybody on the team that has done such hard work to get this out to our customers and our future customers, so thank you very much. With me today is Jasper Grewal, Chief Executive Officer of ROI Technology, a key design partner that has really leveled up our skills in terms of our ability to build plugins for our MCP server. Jasper, why don’t you go ahead and take a minute and introduce yourself, and introduce ROI Technology.

Jasper Grewal: Awesome. Thanks, Rich. Appreciate it. Happy to be here with everyone. My name’s Jasper. I’m the president of ROI Technology. We’re kind of a medium-sized cybersecurity-focused MSP in Washington State, protecting right around 1,000 endpoints. I’ve been working with Syncro to build a couple of cool things on top of their MCP, and I’m really excited to share some of that stuff, and hopefully maybe it’ll spark some ideas in the community about some other cool things that could be built.

Richard Dean: All right, so we’ve got a lot to cover in the time that we have, so let’s go ahead and get started. Quick agenda items. I’m going to quickly talk about what an MCP server is, and we’ll talk a little bit about Syncro’s MCP server. Next, I’ll talk about what a plugin is, and what skills are, and then we’ll talk a little bit about our Service Desk plugin. Then I’m going to hand it over to Jasper, who is going to quickly show you how easy it is to set up the connector, and then we’re going to demonstrate some of the skills that we have co-developed with Jasper. And then we’ll wrap it all up with a Q&A, and get you all the information you need in a place to give us feedback.

What Is an MCP Server?

Richard Dean: So what is an MCP server? Before I get into that, why don’t you, in the chat, give a thumbs up or a thumbs down, so I can kind of get a feel of who here already knows what an MCP server is. That’s old hat to you already. You’ve already built one, you may already be using a few of them. Any thoughts out there?

Jasper Grewal: In comes the flood of thumbs up.

Richard Dean: Yeah, that was kind of the sentiment that I thought the audience would be around, that anybody that signed up for a webinar like this probably already has a pretty good idea what they’re walking into. So I don’t need to spend a lot of time on this slide. I think what’s most important to understand here is that the MCP is an intermediary between your LLM, or your AI clients, and your APIs. It’s the glue in the middle. So instead of writing an API, or writing code to actually connect to the API, get the data, parse it, and then do the next set of things, you put an MCP in the middle, and it’s the glue. The most common analogy that I hear on the internet is the USB-C of development.

Richard Dean: So next, let’s quickly cover what does the Syncro MCP server do today. This is our first version, and with our first version, we have about 40, almost 50 tools at this point that cover tickets, invoicing, customers, assets, appointments, alerts, products, and general search, including document search. We’re investing a lot of time and energy into this, not only around the MCP, extending the MCP around new tools, but also enhancing our API so the experience is much, much better overall from both endpoints.

What Is a Plugin?

Richard Dean: All right, so now let’s get into what is a plugin. Again, thumbs up if you already got this and it’s old hat, or a thumbs down if you’re like, I don’t even know what a plugin is. What’s it looking like, Jasper?

Jasper Grewal: Mostly thumbs up, couple thumbs downs.

Richard Dean: All right, so let’s spend a little bit of time. I think it’s a pretty easy concept to parse together. A plugin is simply a package or a bundle of, first and foremost, skills. That is the primary thing that most of your plugins that you’ll see out there today. They have a set of skills to start, and then they extend from there, because there’s a lot of other things that can go within a plugin. As you can see on the screen, sub-agents and hooks and monitors, and you can look up all of these things. We won’t go into those today. We mostly want to focus on the skills that we wrote, the four skills that we wrote.

The Service Desk Plugin Skills

Richard Dean: So today, our first version of the Service Desk plugin is going to release four skills. I’ll cover two of these, and then Jasper is going to talk a little bit deeper about his contributions.

Call to Ticket simply is a skill for that technician that is on the road, has Claude on their phone, and has maybe had a drive-by or some sort of interaction with a customer, and they need to quickly log a ticket. Maybe they got a call after hours, maybe they’re at a baseball game. Whatever it happens to be, the point of the Call to Ticket skill is to make sure that you didn’t miss any of the information that you should have. Typically you can do this now with our MCP server, you can just ask it to create a ticket with voice or dictation. The problem it sometimes runs into is you miss some of the things. What was the priority? Who was this actually being set to? Did you actually assign it to a tech? Are there full draft notes? Those things. So that’s what this skill is going to remind you of.

The second skill that I want to cover is just the Doc Search skill. It’s simply a grounding skill to make sure that when the MCP is looking for information about Syncro and official documentation, it knows where to look. And then finally, the last two: a research assistant, to basically find the solution that you already have and you just don’t know it. And then the Onsite Visit Prep is simply to avoid overlap of work, duplication of work. So now I will hand it over to Jasper for a full demonstration.

Jasper Grewal: That’s good, we’ll see if the demo gods are happy with us today.

Richard Dean: Yeah, we will see. I chanted, I danced this morning, and I pretty much did all the rituals I could to make sure that we’re good there. But we’ll see.

Live Demo: Connecting Claude to Syncro

Jasper Grewal: All righty, so this is just a demo Syncro account, so there’s not really any real data in here, but there is a lot of seed data, taken from real examples. This is a naked Claude installation, brand new, nothing’s connected, so that’s where I’m at currently. To connect the Syncro MCP, super easy. Just go to extensions, browse, Syncro, connect. So this is Syncro’s OAuth redirect. Say hello. Done. Claude’s connected to Syncro data. Pretty simple. For the plugin, same kind of deal, it’s very straightforward, we’ll just upload a zip.

Richard Dean: And I’ll add here that soon you will have a GitLab repository, you’ll be able to see it in the marketplace, you’ll be able to install and distribute the plugin much easier through just command line if you’re working in Claude Code, so that is coming. Right now, the plugin is available within our Doc Center for a download.

Jasper Grewal: And that’s it. Claude’s now connected to the Syncro data of this tenant, and the plugin with all the skills are loaded, and it’s ready to go.

Live Demo: Call to Ticket

Jasper Grewal: So we’ll do something simple. Let’s grab an organization. Maybe, like Rich said, maybe you’re at a baseball game, or you’re driving, whatever. It’s smart enough to realize what I’m asking it to do. So it found the skill. It’s going to try to find the client in Syncro.

Richard Dean: That’s the fun thing about these type of demos with LLMs, is like, well, let’s watch the LLM think about what it’s going to do.

Jasper Grewal: It works 300 times in testing, and then you try to do it live in front of 200 people, and it’s like, what do you want me to do? I don’t know.

Richard Dean: I’m going to totally do it differently this time. The last six times were just me messing with you.

Jasper Grewal: We’ll get Opus 6, that’ll try to email them or something, I don’t know. No, it’s been really consistent in all of my testing.

Richard Dean: That raises a great question. What model are you consistently running with these types of skills?

Jasper Grewal: So I’ve tested them both with Sonnet and Opus, whatever the latest version at the time was. I think when we started, it was Opus 4.8, and now it’s 5. Fable and Opus should give you pretty similar results. There we go. So now it’s looked at the assets for that company, tried to figure out which one he’s talking about. I’ll say he didn’t say.

Richard Dean: So how well is your organization consuming, are you not going over your consumption level? I’m curious about that. I think when we first started adopting Claude, there was some hiccups for some people. Of, like, oops, that was 200 bucks.

Jasper Grewal: So this account is just on the monthly Pro account, it’s not on an API or a specific consumption account.

Richard Dean: Gotcha.

Jasper Grewal: Let’s get finished. Here we go. 4322, let’s see if that ticket got created. Yeah, so it found that there’s already a ticket open for that customer, that contact. There’s already something here, so let’s try something different.

Richard Dean: No, I’m just trying to wrap my head around this. When you prompted earlier, you were prompting about a ticket, and you wanted a possible resolution for this ticket, right?

Jasper Grewal: No. This is Call to Ticket. We’ll do Ticket Researcher next.

Richard Dean: Okay. That’s what I’m trying to frame in the audience’s brain, so they know where we’re coming from.

Jasper Grewal: Yeah, so Claude thinks that there’s already a ticket that exists that is the one that this contact is calling about. So maybe we look at this, and we’re like, no, this is new, it’s not related to this ticket that I know is already open. So it’s going to prompt for first-time tool use, since this is a brand new connection. I’m going to say always allow, but you might want to be a little more strict with it, so you get comfortable.

Richard Dean: And that’s fully controllable by the administrator within your enterprise account, those types of things.

Jasper Grewal: Teams account too, not just Enterprise. So yeah, this is Pro, kind of like the first entry to paid. Teams is the next step, and then Enterprise, they’re like API consumption-based. Ticket 4541. Throw in some information. Kind of whatever you give it, right? Data in, data out.

Live Demo: Ticket Research Assistant

Jasper Grewal: So you asked about Ticket Researcher, you kind of hinted at it a little bit, so let’s grab an option for that. So maybe I’m a level 1 tech, and I’ve been working with this particular customer, and now I need some help on this ticket. Give it nothing. Bare minimum, help me with this, and your ticket number. So it found the researcher skill. It’s going to go read the ticket and investigate the customer. It’s going to ask, has this happened before? I’ll say new issue. It’s going to show you some web searches, so some people are a little more security-minded, they want to know exactly what it’s doing before it does it, so you can have it show you the searches that it’s going to run, or you can just let it go forward. It does sanitize customer name, it does use the model to try to consider if there’s anything potentially sensitive in the searches it’s doing, so it’s not going to send out serial numbers or anything that’s formatted like PCI compliant data, if you hopefully don’t have that in plain text, but if you for some reason do, it’s going to try to sanitize those searches before it does them.

Richard Dean: So you were able, in this skill, to prompt it to ask you these questions to ensure that it was following the right path.

Jasper Grewal: Yep, and we tried to keep it very light, just a couple of very quick questions. Has this happened before, and do you want to see all my web searches before I do them? The idea being that you’ve already tried everything that you can think of, maybe as a lower-level tech, and you want some suggestions. So you can go do the same searches that it’s doing, but it’s going to do 30 searches in the time it takes you to do maybe 2.

Richard Dean: Yeah, because they can do it in parallel, especially if the instructions are built the right way. And the whole point of skills, if we haven’t stressed it before, is standardization, so that you’re not sitting there and iterating. You can go back to do something else and let it iterate, because it has a set of instructions, and then the fact that your team has the same set of instructions to share is the real purpose and power behind plugins and skills like these.

Jasper Grewal: It’s consistency. Let’s look at the stream of consciousness.

Richard Dean: Yeah, so right now it’s going through its search algorithm, I’ll call it, because that’s an easy word for everybody. Instructions, I’ll just keep calling them instructions.

Jasper Grewal: They used the MCP to research the customer, list their tickets to see, is there context that’s available in prior tickets that might help me here? It’s going to list out the assets that are maybe related to that issue, and now it’s going to give us a hit list of, hey, here’s a few things you should try, kind of in this order, and then it’s going to give you its sources, because we like citing our sources here. And then it’ll give you, at the very end, the condensed version of this very verbose output in two private comments that it can inject into the ticket. There we go. So you’ll get a nice research log. It’s going to offer the research log first, because private comments show most recent at the top. So it’s going to throw the research notes first, so that they’re below the “here’s what you should try.”

Richard Dean: So I’m curious, in your typical workflows. Claude, or all these LLMs, they’re verbose until you force them to be less verbose. That is an ongoing problem across the whole world. I’m curious, when you go through that review cycle, that end part where it summarizes and puts what the private notes would be, or public notes would be, is that where you’re focusing your time? As a reviewer of generated content.

Jasper Grewal: I mean, there’s value in reading the entire thing, but when you’ve got your six tickets deep and the phone’s ringing, you don’t really have time to read it super detailed, so that’s why we wanted to really condense it down. This is what matters, this is what’s super important. Here’s how we came to the conclusion, and then here’s four steps that you can try in this order. There still has to be a level of human intelligence that sits between these actions and the devices.

Richard Dean: Yeah, of course.

Jasper Grewal: There we go. So, 4537. Here’s our “try in order.” And here’s our research notes. We don’t have a live system to test it on, but it did attach the asset, if it wasn’t already. Did all the research. That’s Ticket Researcher in a nutshell. The idea being, if you have a lower-level tech, and they’ve already tried some things, and maybe they don’t quite know where to go from here, before they come without a plan, this can help them formulate that plan. Or even if you have a higher level tech that is just kind of pulling their hair out a little bit, maybe this can give them some new ideas, some directions to go.

Richard Dean: So this standardized the search through tickets and all the known data in Syncro, it’s standardized searching out on the web, and then it’s standardized the general possible set of solutions, and then documented them within the ticket itself. Is that a fair summary?

Jasper Grewal: Perfect.

Richard Dean: All right. So I’m curious, and I’m sure the other MSPs listening are curious, what kind of success have you seen? You’ve deployed this to at least some form or fashion in your environment. What’s the tech feedback?

Jasper Grewal: I have people that never hit their usage limits. Here we did two things, let’s see what a base-level Pro account ended up for usage. Ticket research is not something that every tech is going to use for every ticket, that’s not what it’s meant for. It’s meant to get you unstuck, but I’m on the base tier Pro plan, the cheapest paid plan that’s available, and Opus killed 40% of my 5-hour limit, but only 8% of my weekly limit. Higher level plans are going to have higher limits. The Team plan is going to have higher limits than this. But I have people that are on the Team plan and are hitting their limits and getting a little bit of pushback and having to get upgraded, but those are the people that maybe needed a little bit more coaching, because they were trying to use it for every single ticket. I see that as both a positive and a negative, because they’re using it so much that they’re hitting their limit, but also the negative is that maybe we need to coach them a little bit more of, hey, let’s pump the brakes a little bit, let’s use our brains. It’s there if you need it, but let’s try first. Very powerful tool.

Richard Dean: So this is something that you will continue to deploy to your technicians, and I assume you probably already have some refinements that you want to do? Anything on your mind?

Jasper Grewal: We have a couple of beta releases, maybe we can release those through the Community Center.

Richard Dean: Nice, awesome. Did you want to go through another example, or did you want to move on to a different skill?

Jasper Grewal: Let’s do onsite prep, because I think that’s where a lot of people might actually start to sit up and get a little excited.

Live Demo: Onsite Visit Prep

Richard Dean: Yeah, well, first, before you get into it, tell us the story of how this came about.

Jasper Grewal: Sure. When you’re a 1, 2, 3, 4 person MSP, it’s pretty easy to just throw a message into Teams, hey, I’m going to XYZ customer, does anyone need me to do anything while I’m there? When you’re 5, 6, 7 plus, it starts to get a little noisy, and throwing that Teams message doesn’t really accomplish much. Or maybe you’re already on the road, maybe you went to this customer, and now you’re going to swing by there on the way back to pick up some recycle. But there’s potentially tickets from other techs at that same customer, depending on how you’re structured, where you could have knocked out a quick little monitor move and saved a second truck roll. You could have checked in with somebody that was having some frustration, some printer issues, or whatever, and saved a second ticket, saved a second person from having to go back and do work that you were already there and could have accomplished.

So the idea behind this is, before you go to any given customer, if you’re rolling that truck, throw it into Claude, let it look through the full ticket history of that customer, let it look through everyone’s tickets, not just yours. And maybe identify some opportunities where you could two bird, one stone, or maybe some things that nobody considered could be handled quickly on-site. Hey, my phone has some weird static issue on the line, but you’re not going there for that, you’re going there for something completely unrelated, but you could just swap out a patch cable or something. So just trying to make everyone a little bit more efficient with the truck rolls and things they do.

Richard Dean: You obviously must have had this problem more than once within your organization. Now this is a standard process anytime there’s an on-site visit, essentially?

Jasper Grewal: Yep, 100%. Or I’ll give you another good example. Maybe you have a ticket that’s in a waiting or pending status, so you don’t see it every day, but it was waiting for you to go on site, so you had 3 things you needed to do, and you only remembered 2 of them. So just a good way to surface the opportunities for hands-on work at a given client site. Pick a customer for me.

Richard Dean: St. Aldrick Preparatory School?

Jasper Grewal: Great. We’ll try to make it a little harder on that.

Richard Dean: There you go. I, of course, picked a hard one. I should have said Northgate. That’s okay. Who is St. Aldrick? Is that a real saint?

Jasper Grewal: I don’t know. Random. I have no idea.

Richard Dean: Everyone’s Googling now. So did it assign a skill? I wasn’t paying attention.

Jasper Grewal: It didn’t say it, but yeah, it did fine, it just didn’t throw it in the truncated stream of consciousness. So it’s going to give us a “this definitely needs to happen on site,” “this could probably be included,” and “these are completely unrelated.”

Richard Dean: So let’s get a little insight on skill building here. For the right tool to be called, what is it that you have to do in the skill to make sure that that’s happening in your universe? Because I’m sure different people use different language for things, and that matters?

Jasper Grewal: I mean, it’s largely natural language, and the almost kind of funny thing is you can use Claude to write skills for Claude.

Richard Dean: Yep, for sure.

Jasper Grewal: So you can describe what you’re trying to accomplish. There’s native Anthropic plugins for building skills, too. You don’t have to be really skilled in Python or PowerShell or anything else, you can literally just describe, hey, this is my goal, I want this to be a plugin, I want this to be a skill within this plugin.

And I would encourage anyone who’s going to do anything with Claude software development related to design your own framework, decide what your best practices and standards are, and then refine that, and then for every project that you stand up, bootstrap it with your best practices. We have an internal GitHub repository of all of our policies that cover everything from design to data sanitization to how Claude handles agent dispatch, because it can stand up kind of sub-sessions of itself to handle parallel tasks. Decide how you want that to happen, and then be consistent with it every time. At the beginning of every project, bootstrap, and then maybe keep a local copy of that repository on your workstation, whatever you’re working with. We do all our development in VMs, we don’t do it on our production machines, just to keep things extra isolated. And then when that policy updates, when you change something in it, have a rebootstrap policy, so that Claude will update itself and remember, hey, these changes happened, here’s the new policy, this is how we need to be behaving.

Richard Dean: Yeah, this is a pattern that I see across IT organizations that are adopting tools like Claude and ChatGPT as part of their daily operations, is that building out a dataverse, whether it’s in GitLab, or on your own local database, or whatever it happens to be with your company policies, and the data that it needs, is paramount and essential to these things operating with any kind of consistency. So I think it’s good advice. If you’re exploring these things, I think it’s paramount that you’re giving it the right context, and it has the right bootstrap. So what do we get? What are we seeing now?

Jasper Grewal: Let’s see what we get here. So I’m a tech, I ran this, maybe I’m sitting in my car, maybe I ran it before I went there. Two things they require on-site, looks like they have 12 desktops doing PXE boot. You’re going to have to be there for that. But maybe I didn’t know about the entrance security camera. Maybe this is a different tech’s ticket, I’ve never seen it, I knew nothing about it. But now I know, hey, if I’m going there, I can take care of this really quick.

Richard Dean: And it’s 90 days old.

Jasper Grewal: Yeah. A tech doesn’t want to go out there to replace one camera, because they’ve got to climb up on a 10-foot ladder or something, and the customer’s not complaining about it, and maybe your dispatcher didn’t raise it for some reason, maybe it’s in a pending status. Same thing with upper school. Here’s some opportunistic ones. Four laptops failed the image. You’ve got some options for these, maybe could be handled when you’re on-site, maybe not. Two things at this desk, there’s a RAM upgrade. There’s opportunity here, and that’s not to say that you’re going to touch every single one of these every single time, but maybe it starts a conversation, and if it saves you one truck roll, it’s already paid for itself.

Richard Dean: So go back up towards the top. How is this broken down?

Jasper Grewal: So these definitely require hands on site. You can’t relabel a network closet remotely. It’s hard to troubleshoot a device that’s hard-locking remotely. I can’t do calibration remotely. So these are ones where you have to be on-site to handle these tickets. And they may or may not be yours, but at least this is a one-minute read. These are opportunistic, so these are ones that maybe could need some on-site help, and if you’re there, and whatever you were there to do happened relatively quickly, maybe you could take a peek at these. And then it doesn’t look like they had any “definitely doesn’t require on-site” tickets for that site, but you would normally have a third category of unrelated, this is just noise, these are open tickets, maybe it’s just a conversation in the hallway with somebody, hey, you reported that printer issue, and we haven’t been able to get in touch with you, I’m here for this other issue, do you still need help with that?

Richard Dean: Have you seen any terrible hallucinations? That’s what I’m curious about for this, because it’s going through tickets, primarily. It’s searching through tickets to find exactly what known work is still sitting there for this one particular customer. It’s looking through, these ones would be on-site categorized, and these ones would be something else. I’m just wondering, especially in the something else, have you seen it make up anything? Or has it been pretty grounded?

Jasper Grewal: It’s been fairly grounded. The one place where it can kind of struggle is if you have a customer that has multiple locations, and you don’t have the locations tagged as tickets. But it degrades gracefully, and it says, hey, I don’t know if this is for the location you’re going to, but this ticket from this location, or this from that location. All of the skills that we built should degrade fairly gracefully when the data doesn’t give them enough to work with, or if there’s an API outage, or any of the situations where the perfect use case doesn’t exist. They should all still provide something of value.

Richard Dean: So how much success have you seen with this?

Jasper Grewal: We run it before anyone goes anywhere. Just to cut down on the group Teams channel noise. Before you go somewhere, run this, and at least get an idea if there’s anything else you can handle while you’re there.

Richard Dean: Do the techs do anything after this? So they run this, they review it, and then what’s their general workflow there? Just taking notes, updating tickets?

Jasper Grewal: If you’re going there, maybe throw a DM to your coworker and say, hey, I’m going to this customer, do you need me to, I saw this other ticket, do you want me to handle it while I’m there? Do you want me to move this monitor? Do you want me to talk to this person for this stale security camera ticket? The response on this in particular has been very positive. It’s no longer people trying to remember everything or go through every ticket for the customer they’re going to one by one to figure out, through every status.

Richard Dean: And that’s why it doesn’t get done, because that’s never going to happen. Especially when you’re talking maybe hundreds and hundreds of tickets for a particular customer. Have you found any conflict with maybe two techs starting to schedule something for an on-site visit? Does the skill account for that? Like, this tech is scheduled for this, and this other tech ran it, and sees, oh, John is scheduled for this, something to that effect.

Jasper Grewal: Yeah, it is aware of when a ticket is on your calendar, and it’s aware of which tickets are yours and which are not yours. So it will flag when you have, there’s only one user in this account, so I can’t quite demo that, but it will flag that you have this ticket that requires on-site, and so-and-so has this ticket that requires on-site. And the idea being, reach out to your coworker. Hey, I’m going there today, I see you’re going there tomorrow, do you want me to just handle this one there?

Richard Dean: Okay, that’s perfect.

Jasper Grewal: It’s just leaving all the decision-making and intelligence on the human, which has been my AI philosophy from the very first hour I did. AI should enhance people’s capabilities, not try to replace them.

Richard Dean: Yeah, well, we know at this moment it certainly can’t replace much. It speeds some things up, for sure. So do you have any skills you wanted to sneak peek to the audience, or talk about? We have a few minutes.

Sneak Peek: QBR Report Generation

Jasper Grewal: Yeah, let’s, maybe we can ask for a quick thumbs up, thumbs down, if anyone has issues with QBRs and preparing the reports.

Richard Dean: Oh, yeah. Getting some thumbs up, getting some claps.

Jasper Grewal: Okay. Pick another customer.

Richard Dean: Oh, let’s go with Ironwood Construction. Ironwood. It’s a good name. I believe that they can construct things.

Jasper Grewal: Let’s see how this does. QBR, for us, Syncro is our source of truth. We try to make sure that regardless of what the third-party data source is, anything that would matter to a QBR should be in Syncro in one form or another. Whether that’s in a ticket or in a custom field at the organization level. Invoices, we generate our invoices in Syncro.

Richard Dean: That was the first time the demo gods have struck us down. Maybe.

Jasper Grewal: Okay, it was just a little networking hiccup. All right, so how far back do we want to go? We’ll say 3 months. So we try to make sure that Syncro contains the data that we actually care about. We generate our invoices in Syncro before they go off to our AR system, and so the idea here is ask Claude to go massage that data for you. Don’t run 10 reports in Syncro, and then try to put them into your Word template, or Excel templates, or whatever. Just let Claude handle it. This is kind of its perfect use case.

Richard Dean: Sorry, I’m checking the Q&A. I think the QBR, I’m going to interject my two cents, because I talk to a lot of different IT organizations and MSPs, and I would say QBR, when the whole ChatGPT thing blew up, I think QBRs were like the first few conversations I had. It’s like, oh, I’m trying to build a QBR. I’m integrating Syncro data, and I’m looking at emails for that account, and we’re ingesting this, this, and this. I think it’s the very first thing that comes to mind, on how much time and effort it takes to build a QBR and make it well-received and crafted the right way. You can opine on that better, because you actually do it on a day-to-day basis, but from the feedback, I understand that from a time-saving perspective, being able to build a QBR in the right way for your organization, for your customers, can’t be said.

Jasper Grewal: I mean, a QBR should really be happening more often than quarterly. It should really be at least internally a monthly thing. But it also gets treated as this very rigid thing that needs to happen with a lot of pomp and circumstance, and it really should be more of a guide to a conversation. And this isn’t, even in its perfect form, this is not going to replace a full custom bespoke QBR platform or report platform, but this’ll at least get people that maybe aren’t sure where to start, or they have a disjoint process, or maybe they don’t have a process at all. This will at least get them kind of toe in the water and get them started on that path. But don’t treat this as the end-all be-all. This is a guide to a conversation that you should be having.

And you know your customers better, so when we built this, the idea was we wanted to be able to take different sections of data and show it to different customers. You know your customers, our businesses are all built on relationships, and you know what’s going to resonate with certain people and what’s not going to resonate with certain people. Some people know that they have past due invoices, and so you don’t really need to beat them over the head with it, and some of them, the person that signs the check is not the person doing the review with you, and maybe you do need to beat them over the head with it a little bit. So we’ll see. Takes a minute to run here.

Q&A

Richard Dean: While it’s running, Jasper, why don’t we jump into some of these questions we already have mounted. Gary asks a question around, he says, I’ve been thinking about optimizing costs, especially with tasks like processing events, new and updated tickets, assets, etc. It seems scheduling calling the MCP to process new or updated items is costly and has latency. All true. I tested using Syncro Notification Webhook on these items, and having my code pass the item to the AI. Would this save significant cost?

I’ll throw in my two cents. I totally agree. MCP is not always going to be the answer, in particular for pieces of automation. Your API is probably going to be your best bet, particularly if you want a specific set of data, that’s the set of data you want, and you’re going to do something very deterministic with that particular data. You should be using an API, that’s going to be a lot less costly, there’s no tokens involved. The MCP comes in when you’re trying to do anything predictive, anything to do with reasoning, and anything to do with identification, I would say. So from an automation perspective, reduce costs by using an API wisely, knowing where that actually belongs, versus trying to have something connect through an MCP and do other work. Jasper, I’m sure you have probably strong feelings either way.

Jasper Grewal: Yeah, just to echo what you said, the MCP is not going to replace an API. There’s no world where you want to spend the tokens when, if you already have the capability to write a properly formatted API call and ingest the response and do something useful with that, that’s definitely the way to go. The MCP is really just for empowering your techs to do kind of menial tasks that maybe require a little bit more thinking. Thinking is the wrong word, but it’s the hot word right now. It’s making the API approachable to people that don’t have the capability to stand up a headless Linux server and make calls and configure workflows and webhooks and Teams. I mean, theoretically, everyone on this call should be able to do that, but this’ll, the MCP kind of gets you started. It sort of whets your appetite for it a little bit more. Makes it more approachable.

Richard Dean: Awesome. Well, it’s still working. Let me jump into some more questions. I can probably rapid-fire some of these and get them answered for all of our lovely guests.

Can you make it write a script and run it on the fly? Of course you can make Claude or any LLM write a script as you choose. Can you run it on the fly from the MCP? Not yet. The API for Syncro for running scripts has fairly just been released, at least this summer. We need to put some more guardrails in place before we expose it to the MCP server. That’s what I would say about that.

I’m assuming this is similar functionality and steps on using Copilot instead of Claude. Yes, the MCP server itself is a publicly, remotely hosted MCP server. You could go to our Doc Center and look for the MCP instructions, all the URLs, all the instructions you need are there to connect it to Copilot, to connect it to ChatGPT, to connect it to your local LLMs, whatever you need. It’s fully remotely hosted, so it’s accessible by anyone. All you need is your Syncro account.

Jasper Grewal: I was going to say, do you want me to share this now?

Richard Dean: You know, go for it, brother.

Jasper Grewal: Okay. All right, so, little sneak peek, the idea being, I don’t know necessarily what every MSP is going to want to include in this, so we’ll say maybe we want to theme it Dark Theme Forest. If you have a customer logo, throw it in there. And now we can just run through and see, maybe I want some sample text. You can change the content of it. Maybe I don’t want to mention this.

Richard Dean: So you built a full app. You’re actually generating the, I got it.

Jasper Grewal: We built a framework that Claude then populates, and then there’s one step beyond this, so we can say, maybe I know this ticket’s a sore spot, I don’t need to include that, because it’s already been handled at the time, or it’ll be handled before this gets presented. Maybe we don’t have that recommendation. Do include this recommendation. We don’t want to include an open invoices report, whatever. Change it, tweak it, whatever you want. Say generate. We’ll download an HTML. And now I’ve got a QBR report I can hand a customer if I really wanted to.

Richard Dean: How much time do you think that’s saving you? What was it before?

Jasper Grewal: Incalculable.

Richard Dean: I mean, I assume you were doing all of this manually to some degree before.

Jasper Grewal: Yeah, of course.

Richard Dean: And I know you’re a heavy API user. You, of course, are pulling data into your own tools, but I can’t imagine what it was like before.

Jasper Grewal: For sure. Let me kill Claude, because it froze over there. Hard lock on Claude, that’s the first time that’s happened.

Richard Dean: I have another Q&A while I can answer. Jason asks, is the plugin required to use the connector? No, sir. You can use the connector, and you don’t have to use the plugin. You can create your own plugins, your own skills, as we said.

Jasper Grewal: Go ahead. Other questions?

Richard Dean: Nathan has a question about self-hosted LLMs. Like I said earlier, it’s a fully remote MCP, so you should be able to use a local self-hosted LLM to connect.

What type of Claude subscription is required? No Claude subscription type is required. Any Claude subscription should do. Just depends on your business need and your team’s business needs.

Jasper Grewal: This demo account is just basic Pro. At the bottom of the first step of the QBR console is some steps that you might want to just know about before you go into this meeting. We can scroll down a little bit and see, there’s some internal-only notes. The point of that generate is that you can tailor which of these sections you want to include, but these are maybe things that you should be aware of before you go into the meeting. This particular customer has no agreement, no SLA, no recurring billing. That’s something you should probably know about before you go into that. There’s kind of basic accounting reporting. Spend 5 minutes and read this before you go into that meeting, and you’ll feel better prepared, and the conversation will go probably more the direction you want it to go. Or at least that’s the intent.

Richard Dean: Yeah, I think the key was what you said earlier, you’ve got to know your customer. So as long as you know your customer, you know what’s going to drive them, what’s going to resonate with them, and that’s what this is for, right?

Jasper Grewal: For sure.

Richard Dean: Jordan has a question, and I think he’s talking about the onsite ticket skill. He asks, can you ask if it has nearby clients that have open tickets? I don’t know how that would work, if it knows what nearby is.

Jasper Grewal: Yeah, so nearby is the hard qualifier there, and it’s going to depend on the data that you have in Syncro. We did look at trying to do something like that, and we found a lot of pushback, at least at the time, from the model, on what is nearby. You could ask it, you could say 5 miles, you could say 10 miles, but for us, 10 miles in north is way different than 10 miles down south. That could be the difference between 10 minutes or an hour. So yes, you could probably do that now. Let’s see if it works. Do I have any addresses on these customers, is the next question. What was the customer? It was St. Aldrick, right? I do have some addresses. Let’s just see. I don’t actually know. Let’s do this live, right? I am at 90%, because we jammed through some things, but let’s see.

Richard Dean: The demo gods have been very kind to you, Jasper.

Jasper Grewal: Yeah, right?

Jasper Grewal: Nearby customers might also be worth looking into. We’ll see what happens.

Richard Dean: Hey, I got an interesting question out there. Jonathan asks, could we use the MCP integration with HATZ? Jonathan, I’ll tell you, you should be able to. I have seen registrations, so we track the registration of the clients that are connecting to us, of course, and HATZ has been on that list of connections, so I’m assuming people are already doing it, so go for it, Jonathan.

Jasper Grewal: Yeah, we hit our usage limit, but right behind it, so we did guardrail the skill, I thought we had, just for optimization purposes. But if we can figure out how to have it actually look at addresses and apply maybe some institutional knowledge on different geographies, maybe we’ll try to bring that. I would say right now, you’d probably be better off just asking. If you know your customer, asking specifically, I’m also going to this site, and then seeing if the model is smart enough to cross-reference it for you.

Richard Dean: Fair enough. I think we got a few more slides to get through, and then probably we’ll do a couple more questions.

Is it possible to give the MCP access to worksheets? Jason asks this. Jason, yes, it is possible, and we will be doing that. We plan to expose most of the existing endpoints in the order of right priorities for the MCP server, so wikis, worksheets, estimates, all of those other endpoints that aren’t quite there in the first version will eventually come up. When? I can’t say for now. I don’t know. A lot of these questions are repeats, so hopefully everybody got a good answer for what we answered for other people.

Jasper Grewal: I saw some people opining on already hitting the usage limit, and I was using Opus on the base tier subscription, and I ground through generating an entire HTML QBR based on I don’t know how many tickets are in that customer, probably 100 plus. But I’ve also had pretty good results with Sonnet. So you can downgrade to Sonnet. It may take slightly longer, and it may not be quite as good, but it has been, I’d say 90% is good on Sonnet.

Richard Dean: Yeah, I would agree. Depending on how you’re using it, how optimized your prompts are, your skills are, knowing exactly what you want, instead of kind of playing in a sandbox, is probably the mentality that you need to bring to managing token usage in a lot of sense. So that would be my advice. Overall, we are making the API more efficient so that when these calls and the data comes back, we’re not burning as many tokens, so that’s greatly on our mind as well.

Jasper Grewal: And that’s why we bounded onsite prep to be one customer, because the model tries to be helpful. It’s going to try to go out and do a little bit more and gold plate the response. But that’s why we bounded it to one customer.

Where to Find Everything

Richard Dean: All right. So let’s wrap this puppy up so everybody knows where to find everything that they need to know and how to give us some feedback.

So, where to find the Syncro MCP? Like we’ve said, it is in the Claude connector. So you can look for the Claude connector right now, you can read about it, you can request from your administrator access to that connector, so that they properly publish it within your organization. You can search for us anywhere on the web. We’re in multiple MCP registries, and we’ll be adding more. ChatGPT is coming. We’re going to get inside their plugin registry, and of course, always go to the Syncro Doc Center as the authority for everything Syncro. So that’s where I would start if you have any questions.

Installation is super easy, like Jasper showed you in the demonstration. It’s just OAuth authentication. You log in with your account, we accept the permissions that your current account has, and that’s how the MCP is going to work. So if you don’t have access to that endpoint, the MCP is not going to allow you that access. And then simply connect. You’re just signing in.

Jasper Grewal: On that note, just a quick little tip, it’s worth auditing your technician’s permissions before you turn them loose with MCP, just to make sure that people have the permissions that they’re going to need, and also don’t have unused permissions that they don’t need, because Claude tries to be very helpful, and if they have a latent permission that they have never used, maybe they can get to invoices, but they didn’t realize it. Claude’s going to take advantage of that if it can, to try to help them. So just a quick little run through your Security Center, make sure everyone has what they need and doesn’t have anything they don’t.

Richard Dean: Another great tip for all the MSPs and IT orgs out there, so thank you, sir.

Where to find the Claude plugin? Right now, you can find it on the Syncro Doc Center, like I said, but as I previously mentioned, soon it will be out in our own personal verified GitLab repository, and that’s where we will start sharing all of our shareware type of solutions related to the MCP server. So we’ll be announcing that soon.

Of course, within our Doc Center, you’ll also find a prompt library of curated prompts from our different partners. Again, these are prompts to get you started. They’re not all going to work perfectly for your organization, so take them, learn from them, because these are from our best partners, and then use them to your best ability, and hopefully please share them.

We’d love to hear your feedback. You can go to our Community Center and provide any feedback that you’d like on the MCP server, on the plugin, in AI in general. But we’d love to also have you share anything that you’d like to share. Share prompts with each other. So we really encourage that, because I think for MCP, for this type of forward extensibility within your platform, it’s important that we all share with each other, because this is a whole learning process. So please, post to the community.

Closing

Richard Dean: And then finally, if you’re not a Syncro Partner right now, we encourage you to contact one of our representatives to get a demo, or you can try yourself a completely free trial. It’s the full stack, it’s not just part of the platform, so you can try out anything. It’s 14 days. There’s no credit card information that you’re going to have to enter, so none of that type of thing. So please, take a look at Syncro, take a look at our MCP, take a look at our other AI features that we have released and are developing. We’re pushing MSPs and IT forward.

So, thank you for your time today. Jasper, thank you for your time and all your efforts on this, sir. This wouldn’t have been possible without you. And the fact that we did this in the time crunch that we did, plus vacations in between and everything else, a miracle, sir, a miracle. So, thank you.

Jasper Grewal: It was a lot of fun, I appreciate it.

Richard Dean: It’s always a lot of fun, sir. I appreciate all the things you do. So, thank you to everybody, and everybody have a wonderful afternoon, and see you on the next one.

Jasper Grewal: See you later. Bye, everyone.