Debugging and Impact Analysis

Steven Hloros

0:00 Well, welcome everyone. Hello, Megan. As you can tell, I am watching over chat, so I am here today, but would like to welcome everyone to our first of three Summer of Atlan MCP events. These are going to be quick, 30-minute kind of deep dives into different personas that we really see, you know, our community utilizing the MCP to accelerate their processes and really get that context from Atlan in the tools that they're already using.

0:30 So today, our agenda is we're gonna be deep diving a little bit into the data engineering persona, taking a look at what we would think a, a day in the life might be and the questions that usually pile up for that poor data engineer. We're then gonna look at four different demos, all based in actual conversations with some of our community members in how they're using the Atlan MCP with other tools, in this case, Claude Code, to not only solve their issues, but kind of connect Jira directly to it or your ticketing system.

1:04 We're then finally gonna look at what's next with the Atlan MCP and a couple calls to action to make sure you're coming out of this with the actual, you know things to utilize it for. How can you kind of bring this to your larger group? So with that, I want to ask a quick question to all of you. We have seen a lot of uptick when it comes to the Atlan MCP and really bringing that context into the tools that you're actually using. But we wanted to ask the community that's here today to be honest in what's actually kept the MCP maybe from catching on with your company.

1:43 So I'm gonna launch this in the Slack poll, but there are a couple different options. And we really wanna hear from you. Is it that you haven't prioritized it over the Atlan AI that you're already UI that you're already using? Is it just people are so ingrained in their, their ways, they... It's not worth the time, they think, to actually start using the MCP? Do you not know what you'd actually use it for? Are the use cases just not kind of piling up for you, "Hey, this would be a really great way to utilize it"? Are you held up by security and IT? Is there a review that goes on?

2:17 Are you just waiting for that final sign-off? Did you realize it was even available to you? By the way, it's available for every Atlan tenant, so your tenant already has it. You can literally connect today. Did it fall short for you? So have you been using it, but it just never really caught on, maybe because of token usage, or maybe you just didn't know how to use it best? Or finally, has it already caught on for you? You're already rolling it out. So I'm gonna give this just a moment here and see where this comes in. So majority is haven't prioritized it over the Atlan UI we already use.

2:52 Some of you also with the not sure what you'd actually use it for. Four, 15% are held up by security and IT. That's, that's very interesting. Also, we have security docs, so if you ever need anything for security, we absolutely have those in documentation. Hopefully, hopefully it's just more what your company actually has to go to. A few of you didn't realize it was even available, and several of you have already caught on, about three three different participants, about 11%. Okay, great. So it's really about prioritization. That's very interesting to know. Awesome. Thank you so much for taking part in that poll.

3:30 This really helps us because this summer we are really looking to, you know, deliver on the benefit of the Atlan MCP, really get people utilizing it in the tools they're already in. With everything that's going on with productivity, with AI, the Atlan MCP really can be that boost that your data teams, your data engineers, your business users really can use without having to onboard them onto Atlan and understand the UI. That's the, the beauty of something like the Atlan MCP. It doesn't really take a lot of training. It's just using your actual conversation to access your context. I'm gonna go ahead and end that, and we're gonna go ahead and start.

4:12 So thank you all so much for posting in there. To ground us, just what is an MCP and what is the Atlan MCP? The MCP, it's just the open standard that lets your AI tools that you already use communicate with another system. So whether you're using Claude, Cursor, Copilot, ChatGPT, whatever they are, an MCP essentially connects it to different tools. So we have an Atlan MCP, and that gives your AI tool the understanding. It gives them tools to actually interact with the source system. So the Atlan MCP has all the instructions on how to do things like searching and updating metadata.

4:52 And so your actual tool, Claude, Cursor, whatever it may be, you don't have to know the APIs. You don't know how, have to know how Atlan actually works. You just in plain language say things like, "Can you find me any assets that have MRR?" And the Atlan MCP actually translates that with tools to interact with Atlan, bring that context right back. So you don't have to be technical. You don't have to understand how to actually do authentication inside of Atlan. The Atlan MCP covers all of that. And for those of you that don't know, the Atlan MCP can do a lot. There are two parts to it. There's consumption and updates.

5:31 So you have the ability to do things like search and discover and find your assets. You can look through lineage and understand impact and root cause analysis. You can also do things like enrich metadata, add descriptions, add tags, whatever it may be. You can also govern and certify. You can set certifications and ownership. You can create glossary terms. And finally, if you have things like Data Quality Studio with Atlan, you can actually create data quality signals right from Claude right back to Atlan, and those will get created there. And you can even consume those quality signals with the Atlan MCP.

6:04 So the Atlan MCP itself is a whole tool belt that is given to your AI agent.

6:11 For today, though, we are focused on the data engineering persona. In that data engineering persona, I always see this meme. I'm sure you've seen it on whether it's in LinkedIn, Reddit, whatever it may be. But anyone that works with data, they always have that 2:00 AM question. And for data engineers, a lot of the times are it's, "Hey, if I made this change that was wanted, what's actually gonna break? And if I made that change already, is something breaking?" Or someone asks why this dashboard is broken, and I didn't get that. Why is it still broken? It's what keeps data engineers up at night.

6:47 And those two questions are what eat the time from a data engineer's week. And so today, we're gonna be focusing not only on those two items, but also two additional ones and how the Atlan MCP can actually give them all the context they need to resolve these type of tickets without leaving their AI agent. So we see a lot of times in talking with a lot of our community for those, those data engineers, that there are four different types of questions that really do come up. The top two are the ones that always come up. Hey, a dashboard's broken. Hey, a pipeline's broken. Why is it broken? What's going on?

7:25 What's the root cause of this? And the second one is they get the request, "Hey, we need to change the calculation of this column or this metric." And before they do this, they need to know, well, what actually breaks? And a lot of the times, that's a very long process, and sometimes those data engineers will actually skip that entire process and just make the change, which then keeps them up at 2:00 AM. We also hear often that the question of how is something calculated comes to a data engineer's plate because they're the ones who actually modeled the data. They created, let's say, an MRR field with the calculation.

Steven Hloros

8:00 And finally, they get the request of, "Hey, I need a new MRR calculation. Go ahead and create that view for me." And a lot of the times, that might also exist already, and so it creates duplicative work. And all of these are things that normally land inside of a data engineer's, whether it's a Jira inbox, whether it's a task inbox. Wherever they work, they all land there, and their day starts going through first the fires and then going through the additional questions. And so here today, we're actually going to go from ticket to answer. We're not gonna do any slides here. We're gonna work directly inside of Claude Code as a data engineer likely would.

8:43 So I'm gonna put on my data engineer hat for this one. And as a now data engineer, I'm gonna go ahead and start my day. And so I might log into Claude Code. And here inside of Claude Code, I have both the Atlan MCP and the MCP to connect to Jira all set up. And so my day might start with a very simplistic question, and that's, "Hey, what tickets are actually on my desk?" And in this case, I created a simple skill. People always create skills to know how to prioritize, how to actually get those. So here, just asking simply, "Hey, can I get a triage of my open Jira tickets?"

9:21 As a lot of our data engineers or other personas will do if they're using a ticketing system. So here we can see it actually goes in. It's using the Atlassian MCP to pull open my Jira tickets. So as it calls this, we're gonna actually see, hey, for this, we have four different problems that if we go back to the slide, we alluded to before. And those are, first, we have a ticket asking about an issue with the worldwide importers overview Power BI report. There are blanks, it's under-counting, and unfortunately, stakeholders actually don't trust it, and there's a Friday business review coming up.

10:01 Second, finance actually wants the tax on fact orders recalculated with the new regional rate for FY twenty-six, the financial year twenty-six, and they need a go, no-go before any changes are made or before I make any changes. Next, FP&A keeps asking how MRR is calculated for their board reporting. They need to understand it so they can explain it. And finally, there's a new exec view needed that uses MRR, and the question is, do we need to rebuild this, or can we reuse it? And as I said, this might be the normal day.

10:38 You start off with all of the different tickets that you need, and instead of traversing back and forth, finding the assets, looking inside of Atlan, looking inside of different tools, with the Atlan MCP connected, we can actually triage all of these tickets right from Claude Code. So in this case, it actually gives a recommendation. Again, I have built a triage skill for myself, so I know what's most important. We can see there's one that's on fire. And usually anything that is broken is always the one that's on fire. And in this case, that is about the worldwide importers overview Power BI report. There are blanks, and it is critical for Friday's business review.

11:18 So in this case, I'm just gonna go ahead and tell Claude, "Hey, yeah, let's go ahead and start with DCTN-2," our ticket for that. I wanna look upstream of the worldwide importers overview Power BI report to identify if there are any issues upstream inside of Databricks that may be causing this issue, right? A data engineer hears, hey, there's an issue, something's broken. What's the first thing to do? It's to try and look upstream to see if there's an actual root cause for that break. So in this case, gonna go ahead and enter that.

11:51 And what this is going to do, because I have the Atlan MCP connected, it's going to actually resolve that back in Atlan. So it's actually calling Atlan in this case, which is hooked up to my instance, and it's going to trace upstream. And I, of course, always works up until you are right back inside authenticate Apologize on that.

12:14 Was working five minutes ago. It always does that to me. And let's go again. All right, we are authenticated again, and we're going to resolve the report now upstream. So it's gonna call Atlan in this case. First, it's going to actually search for that worldwide importer's overview Power BI report, and we can see it's actually using semantic search to find it. So it's taking what you would normally do inside of Atlan from the UI and try to find that asset. So if I were actually using the Atlan UI, that might consist of me actually going inside of Atlan. Let me move this.

12:55 And either using our conversational search, or if I knew exactly where to look, I could say worldwide importers to find that asset. And again, here, absolutely great to take a look at, try to find, but in this case, with the MCP, I don't have to switch context. And so what we see here is first it found that Power BI report, and it's actually even telling us that it's owned by operations. Next, it's doing the tracing upstream. The upstream trace lands on the Databricks gold layer, and it finds that one Databricks table carries an issue flagged on it, the fact orders table.

Steven Hloros

13:34 It actually goes into Atlan and gives us exactly what's happening on that upstream table. And we can see here there's actually an announcement saying that the daily data quality checks today flagged that the customer ID null rate was over fourteen%, up from the less than one percent baseline, and the platform team is already investigating this. But it's telling us, "Do not trust the customer level cuts on dependent dashboards until this clears." So here, right in one simple call, I've been able to take that dashboard that is broken, the ticket that I've gotten, traverse upstream to find an asset that has been flagged, in this case with an announcement.

14:12 But if it had Data Quality Studio or a connection to Soda or Monte Carlo with failures, the Atlan MCP would bring that information back and actually give me the understanding of, hey, this is the issue for that dashboard. And what's amazing here is it not only gives me that understanding, but kind of contextualizes it. It tells me that the sales group actually owns the, the fact orders, but the platform team is actively working on it. Now I know exactly who to go out to, and the great thing about an MCP is, here, I could literally update that Jira ticket because, again, I have the MCP connection.

14:50 So I could literally type back to write a summary, post it there that, "Hey, in the next twenty-four hours, this should be cleaned up. If you don't see that that's been cleaned up, please do not actively use that dashboard. It won't be ready." This is a way, instead of, again, if I were to go back into Atlan as a data engineer, find first that Power BI report, go into Lineage, and traverse all the lineage through and through.

15:18 Instead of doing all of that work from the Power BI report back in the order fact tables and getting the actual announcement that's on this asset itself, we can see here the exact announcement that the MCP pulled back. I get all of that right inside of Claude. Literally everything that might have taken me fifteen minutes to actually traverse backstream, all done right inside Claude.

15:42 And that's what we've seen a lot of our community members doing, is the jobs that they normally do within Atlan that are repetitive, that are just basic, "Hey, I need to traverse that lineage upstream," you can do right inside the MCP without having to use the UI, download an Excel file, look through all that information. So in this case, I now have my answer. Hey, there's an upstream broken table. It'll be resolved in twenty-four days. I know exactly who to reach out to and contact if I need an update. I know exactly from the requester perspective who to actually guide them to and why it's broken. And so that's the first ticket solved.

16:19 But I have four tickets. So now I I'll say what are the remaining open tickets? And again, because I'm connected to Jira in this case, it's gonna go back. It's gonna find, hey, from that session, we've already completed our first task, which is the customer accounts was wrong on the WWI overview report. Found that root cause of it all. Next, we have an actual change. Again, what we see quite often from data engineers or anybody that is actually building the data, and that is, hey, we want a change to the underlying data. We want a different calculation. Well, when you make those changes, again, it might impact others that you weren't expecting.

17:04 And a lot of the times, that's a very long process to understand the impact analysis, even from the UI itself. But with the Atlan MCP, very simple. So again, here, I'm just going to let it know, yeah, I want to investigate this. Before I make any changes on what they requested, what's the actual downstream blast radius? And are there any production reports that may be impacted? And who actually owns those? Because before I make any of those changes, I wanna make sure I'm reaching out and understanding the impact that it may have.

17:37 Before I make that decision, before it's, "Hey, I can update this table," or maybe I need to create a different source for that table with the new calculation. I need to first interview the stakeholders to make sure, hey, on Friday, I'm not gonna break that important dashboard. So I'm gonna here just ask it, before I make those changes, what's the downstream blast radius, and are any of the production reports impacted?

Steven Hloros

18:01 Here, it's gonna go ahead. It's gonna find that fact orders table, and per our transformation, it's gonna resolve down the Databricks copy all the way down and trace where it has dependencies downstream. Again, it's going to call Atlan in this case. Again, Atlan MCP is connected, and we get a decision here, again, right inside of Claude. Changing the tax rate on the fact order propagates through two BI stacks. So we actually have two different ones that are using this. Collapsing orders into duplicates. There are six production BI consumer reports across Power BI and Tableau. And here's the key, the ones that matter for a tax change.

18:43 We see immediately import, import/export performance is a verified, certified dashboard that utilizes this. This is a trusted asset. So before I make any changes, I already know there is a dashboard that is certified, that is verified, that people are depending on, and that I need to talk with those owners right away before I even consider making any changes. And here we see exactly who to notify. Again, getting all that context from Atlan and telling us directly inside of here, "Hey, who actually owns those reports? Who do I need to reach out to?" And it gives me even a verdict in this case. I need to coordinate first, not go free.

19:28 I need to talk with those actual owners, make sure if we make the change, will it break anything? And the cool thing here is, again, if we were to consider this from the Atlan UI perspective, here's my actual asset that I was working with, my order facts table, and I have a column in there for tax rate that may be modified. As a data engineer, normally to do this impact analysis, I'd have to walk through the lineage one by one, check each one of these, see who actually utilizes it, see who the owners are, so on and so forth.

19:59 Or, again, inside the UI, I could do an impact report, and I could export it to Google Sheets. But then I'm playing with a CSV, I'm trying to find all that information. With the Atlan MCP, makes my life extremely simple. It tells me the exact assets that are downstream of it, who may be impacted, and even the ownership to actually follow up with them. So again, now I know exactly who to work with, and I haven't had to switch context. It's all about bringing that context right into the tool that you are using. And so in this case, hey, I know exactly who to notify. I know that I can't make a change today.

20:36 I need to go back to the person that created that ticket, have a conversation with them to let them know, "Can't make it just yet," but we will get there, and probably schedule a call with the, the ownership team of these different BI reports to make sure if we make this change, will it be okay? And if not, maybe create a brand-new view that would allow for this tax change and work with that stakeholder so that they can utilize it. So now I've resolved this ticket. I know exactly who to reach out to, and I can say, "What are my remaining open tickets?"

21:09 Again, because I'm connected to Jira through the MCP, it's gonna bring back exactly what I've done. So we can see, hey, I've addressed the two biggest tickets already. We have the one when it came to the root cause. I finished the second one, impact assess. I know exactly who I need to reach out to. Now the third one, which is confirm the official MRR definition. And again, since I'm in the MCP, it makes it easy, but I could get more details. If I look at my board, it's pulling directly from those Jira tickets that I have. So we see here, how is monthly recurring revenue calculated? That's the ticket. I need to answer that.

21:45 So I can simply ask, "How is this actually documented?" I wanna go ahead, move forward with the DCTN-5, and it's a simple definition lookup. Again, the Atlan MCP knows how to get the context from Atlan. So I'm asking for the MRR. It's gonna check our glossary, the terms that I have, and try and actually find that MRR glossary term in this case, and see if it can't bring back all the information about how it's actually calculated. Here we see it's found an actual governed MRR term that's certified. It's owned by finance, and it actually will pull the full documentation to service the calculation.

22:29 If I were to go into Atlan, what we'd see is a normal Atlan term with MRR. But what's cool here is the Atlan MCP isn't just about the, you know, finding the context around the name of it, but it's gonna provide me with all the information around that exact term. So we can see it's a verified glossary term, but it's gonna give me all sorts of additional information from the official definition to, if I wanted, is there a direct calculation on the term for MRR?

Steven Hloros

23:04 Which, hey, can I find the exact definition, the exact calculation, and it's gonna pull the README. A lot of our customers in their, our community actually put in things like full calculations inside of the README. So in this case, might have asked that. The README has an overview on the MRR. I think my, my little part think it got erased, and I apologize that.

Steven Hloros

23:31 But if it did have the exact SQL there, it would actually be able to pull that back. And this is a way, again, as a data engineer, when we get these one-off questions, "How is this actually calculated?" If it's documented inside of Atlan, it's a very simplistic way to go ahead, get that definition, and pass it back. But it's also something when you think of business users, they should be able to do themselves.

23:54 So actually for this one as a data engineer, I'd provide the description, and I'd provide the calculation if it was fully documented, but I'd also tell my requester, "Hey, this is something you could actually pull yourself simply by using the Atlan MCP." So with— My final ticket. So I now have the definition. I can send that off. It has how it was calculated. If we look back at the definition, it's revenue for each month based on current subscription, excludes one-time charges. So we have exactly how and what is involved with that. And finally, we have the new exec view MRR ticket. So somebody's asking for an actual view that has that MRR information, that calculation.

24:38 But before building it from scratch, again, the context is all inside of Atlan. And because our team has tagged different data assets and information there, I can actually find which assets already have that information. So I'm gonna go on to number six, and before I build MRR into a new view, does it already exist? Does the data already exist somewhere? Again, if I was in the Atlan UI, I might go into the search. I might search for MRR in this case, try to find the term, try to find exactly what it's associated with, which columns, which data assets.

25:13 But here with the Atlan MCP, tells me exactly, hey, it's actually already connected to the executive summary total monthly recurring revenue. And it pulls the information. It kind of gives all the information from that ticket because they were actually asking about an executive summary, and it's able to match on a like data asset that's specifically for executive summary. Their ticket asks for that. We're getting the total monthly recurring revenue. We already have it. So inside of Claude, because it used the Atlan MCP and got that context, it tells us, "Hey, we already have it." There's no need to rewrite SQL in this case.

25:51 Point the requester exactly at this table and see if it fits their needs. Now, it might not. They might need something different. But at the very least, we're giving an answer right away. We're pointing to them to something that already exists, and it saves us time from having to do all the manual searching and finding. And so we can see here, it's a governed exec grade MRR already tied to same term. In just simple couple minutes here, what we've done is we've gone through four different tickets as a data engineer. Throughout all of it, we stayed inside of Claude Code, the tool they may already be using. They might be coding inside of there.

26:33 They might have it connected to the Snowflake MCP, and they're bringing in different data or they're building new tables. It's all there from them. In that root cause, it was brought right to them through Atlan without having to traverse the lineage. It's all the same governed context right inside the tool. And the thing I wanna point out here is these are just four quick available options that literally tomorrow any of your data engineers could start using Atlan for.

27:02 Even if you don't have, in this case, if you don't have things like maybe ownership or you haven't fully built out all the descriptions that you want, at the end of the day, that lineage is auto-generated by Atlan. So at the very least, they can traverse the lineage to see what's actually feeding dashboards or reports and get clued into exactly where they should work. And this same context isn't just for a single consumption. It's something that we're seeing community members build upon. So if you have things like internal chat bots that everybody uses, you can connect those to the Atlan MCP so that when someone asks, "What is MRR?"

27:42 Or they wanna refer to standard definitions, it's all there for them. Now, very quickly, I know we only have a few minutes left. I saw a couple people are actually rolling out the Atlan MCP. And the thing I wanted to point out, and we're gonna do this throughout each of these, is updates with the Atlan MCP. But I wanted to point out the Atlan MCP actually just got better. I'm sure you watched it, and yes, I had skills, and obviously it's a demo. But in, in thinking, "Hey, this went a little too fast," yes, it is a demo. It's fictional data.

28:14 But the great news here is this is also a result of what we've been working on, which is making the Atlan MCP better. We've been working very hard to make it faster, smarter, and leaner, costing less tokens, making sure that it's doing better searches. We've also added in access controls for you. So if that was something that you were worried about, maybe you only wanted your administrators to use the MCP to start with, that's now available to you. So if you're an admin, we actually have inside of our Labs section now the ability for you to turn on and off the Atlan MCP, and you can specify whether it's only admins or all users.

28:54 Our goal is to make the Atlan MCP something that you can hook up, have ready to go, and feel confident in as quickly as possible. So as next steps, as I talked about, the things I would ask you all to do. Number one, connect the Atlan MCP. If you haven't already, it's unbelievably simple now. I'm gonna post this in chat. But essentially, inside of here, you can pick the tool of your choice, and it's going to walk you through exactly how to set it up. In Claude Code, it's as simple as a plugin. No additional setup needed. You don't need to ask for it to actually be turned on for your instance. It's already there.

29:33 And you get taken to a very simplistic auth page, and it uses your authorization that you already have inside of Atlan. Work your first ticket. If you're a data engineer, go ahead and try it out. If you're doing root cause or impact analysis, if you get asked by it about a definition, try it out in the Atlan MCP. And then third, send the recording to your team. We'll be following up with an actual recording of this.

29:55 But if you have data engineers, if you have analysts, if you have anybody that's using Atlan right now, send this off to them as different ideas for them to actually start using it, maybe keeping them from switching context back and forth and changing windows all the time, making them get the context when they need it.

30:14 And finally, just as a heads up, what's happening this summer? Well, first, we have our second session, which will be on business analyst persona. That'll be next week, same, same place, same time. And we'll be going over a little bit about possibly how you could build an actual dashboard from the context inside of Atlan, along with answering data questions. Also, we're gonna have a new refer and earn raffle coming up in August. Be on the lookout for that. But for new users of the Atlan MCP, we're gonna have a refer and earn program where you'll be entered into a raffle. And finally, we're gonna be publishing the Atlan MCP Cookbook.

30:56 This is going to start off with actual Atlan recipes, such as what you saw here, but we want this to be something, as time goes on, that our community builds out. So be on the lookout for that. We'll be asking for submissions and different, different uses of the Atlan MCP. With that, I'm gonna go ahead. I will say goodbye. One question in chat, if I Well, is it possible to start testing just admin users? A hundred percent. All you have to do, this literally came out last week. If you're an admin, you go into Labs, you go into Atlan MCP, and you can say, "Only admins have access to the Atlan MCP."

31:35 So only admins would be able to do it. The thing to point out to everybody is when it comes to the Atlan MCP, uses a OAuth connection, so if a person has certain permissions, those traverse to the actual MCP. But to Sabrina's point, if you're looking to just test it out, get your admins in there, see how it is, we have now released inside of Labs, you'll need to be an administrator inside of Atlan. You'll see Atlan AI in an Atlan MCP. You can set that to all only admins or all users. All right, I know it was quick, I know it was fast, but hopefully that gave you some ideas.

32:12 We'll be running another one on Thursday next week, same time, same place. But thank you all so much. A recording will be sent to you tomorrow, and I hope you all have a wonderful rest of the week. Have a wonderful summer. Stay out of the heat, and I will see you all hopefully next week and hope to start hearing some success stories. Thanks, everyone