The MCP Sessions Vol. 1: Sports Analytics

0:00All right, hope everyone enjoyed that intro. Uh, I am Jacob over here on the Devril side at Motherduck. We are super super excited to have you here today for our first of the MCP sessions. And um, I

0:14just want to quickly hand it over to Logan here. Logan, why don't you introduce yourself?

0:19 >> Hey, I'm Logan. Uh, I'm in Seattle based on GoHawks. You can see the vest. Uh, and I'm on the accounts team here at Motherdoc. And clearly I'm like Madonna.

0:28I guess I just have one name. Jacob, you uh might want to drop the last name at some point.

0:34 >> Yeah, right. Maybe I do need to do that. You know, actually uh at Motherduck, I go by my last name because there was a another Jacob here when I started.

0:43 >> So, uh there's some there's some lore there on my email. Um for those of you that don't know, I will uh I will Logan was also a professional golfer in another life. And so we get to do lots of fun sports analytics later with a domain expert, which is what is one of my favorite things about what we get to

1:01do here. So uh I am going to uh add

1:06something to the stage here. It's clotting coffee time. This is what we're doing. We're going to be doing some analysis, some vibe analysis. I want to just show you real quickly everything that you need in order to do this type of stuff yourself. So, the first thing I want to show you is I'm just going to go

1:23into my settings here and I'm going to hop into capabilities. You'll see that I have actually added a um way to generate charts using uh JSON specs. We'll touch about that later.

1:36It's called Invis. I'll hopefully be open sourcing that uh later today. And then I want to show you the connector.

1:42This is where the magic happens. We've got the mother duck connector. This just went on the Anthropic approved connector list uh yesterday. So they gave us a gift just for this live stream and you can see it is now in here which means you can add it to your cloud. It is part of the approved kind of marketplace

2:02adapters. It is our remote MCP. Um really great tool for interacting with the database. Uh really really excited to show it off today. All right, let's start a new chat. Um, so Logan and I actually did some exploration last week looking at um playoff odds and I run a um

2:26the Monte Carlo simulation that simulates kind of NFL season. And so we're going to actually use that data set. Um let me go grab my prompt here.

2:36Let me hop into my chat where I saved it

2:41and I'm going to send it over. Hey, why does it have a little You know what?

2:44Let's fix this. I don't want to that. Okay. So, here's my [clears throat] prompt. Uh I want to do some analysis on NFL data and motherduck. I have that hint in there to tell it, hey, maybe I need to use motherduck MTP. You can use without that and maybe it's smart enough to know that you want to use the MTP.

2:59But in general, it hints for tools are very helpful to get kind of the ball rolling. So, I like to do that. And then I tell it, hey, there's two databases.

3:05There's this one MDSbox copy, which is the data after week 18, and coffee WC, which is after the wild card weekend.

3:12Last game was last night, right? Um, we saw the Texans take down the Steelers.

3:17So, let's take a look at it. It's just going to diff these two databases, right? One of them has is end of the regular season. One of them is after the wild card game. So, there's going to be some different changes in simulation based on what happened there. So, let's run it.

3:30Now, this is the fun part where we get to wait around. Logan, how much coffee you have had today?

3:35 >> Uh, just one cup today. >> Okay. Sun's out, right? So, >> yeah, that was fair. [laughter] >> Exactly. before. I typically do one before lunch, one after lunch. That's >> okay. Great. [laughter] Amazing. My coffee increases way super high in in the winter, but not on sunny days. It's really about the uh >> it is nice.

3:53 >> So [laughter] indeed, let's um let's take a look at what tools are actually being called here. So mother duck the motherduck MCP has a set of tools that um are used by claude. And so we can actually look at the call. So the first one is list tables. And so it says, hey, for database with this name, which I gave it

4:10up here, it says, what is in here? And so here's all of our tables, right? It returns a JSON response. And then it's doing the same thing for our other database, right? So it's like, all right, cool. Um, and it says, I see the wild card snapshot was added has added additional view. That's true. I did add

4:25that view. Um, uh, with the help of Claude last night, we added these changes to my model. Um, which is awesome. And so now once it has the tables, it can start writing queries.

4:35You'll see this first one didn't like uh gave us gave it an error uh reference the column not found in the from clause.

4:41So, oh let me recheck it's going to recheck the um uh it'll recheck the columns and be like oh I see what I to join this on. So it's looking at these this raw results table >> um which has the actual data just coming right in from uh from in pro football

5:01reference actually and then we write some queries right. Okay, how many uh how many games are there in this data set in both databases? So, you can see this one gets 278.

5:14This one gets I think 272. So, it immediately is like, hey, we've got six additional gangs. What games were added and it is basically kind of stepping through just like it's an analyst um and seeing what's going on.

5:26 >> This is my favorite thing though is like while you're like interesting wild card results. It's like I'm already getting excited. I'm like, "Oh, yeah, fascinating." >> Totally. Totally. I love um I love the enthusiasm it brings. Like it really feels like when you're using especially like Opus 45 that it's very curious um as it's diving through

5:44 >> uh diving through the data. So, um you can see it's continuing to write queries here. What's great about SQL is that um uh the syntax because the syntax is validated and it gives good error messages back, it's pretty easy for it to write it. Um and duct DB of course is mostly um Postgress compatible and so

6:03there's lots of training data for it to write these queries. All right. So it's continued to work on this. Um it has noticed something interesting which is um in my model I do not um I've not added a feature to change the ELO ratings for teams once the playoffs start. And I haven't really thought about how I want to implement that yet.

6:24So maybe in the future I'll make it so that you know depending on playoff outcomes ELO ratings change. Um for now they don't which it has detected and is true.

6:32 >> Yeah. Essentially saying you are the team you are once you enter the playoffs.

6:38 >> I mean I think uh I the truth is I actually haven't read the the papers on how to do this um for the playoffs in the NFL. My model was built for regular season simulation, right? Um, and playoff. I just added playoffs this year and so I haven't done any back testing or tried methodologies that are

6:56effective for building a good model. >> Yeah. >> Um, they might be the same, but that is exactly right, Logan. You're correct. It is doing that. All right. So, it's continuing to do this. Um, one thing that's interesting is it's notice it says these views have catalog reference issue. That's because I copied them from a local duct DB file into motherduck

7:15with a different name and the viewames in the underlying tables reference the old um reference the original name of the database. Um that's not a problem because what we can actually do well a because everything's materialized that we care about is materialized as tables.

7:30But we can actually also tell the model something like hey um these views are actually syntactically valid. They just need to change the from clause to the right database name. And you can look at the information schema to find out how to do that. All right. So, it's built a first analysis. Let's take a look here.

7:47Is it right? >> Yes. So far, >> that is correct. >> Let's see. Yeah, I really wanted the Panthers to win.

7:55 >> I'm really I'm still disappointed. as your as both your model before the round stated and as my own eyes said the team most feared is the Rams when I I'm a Seahawks fan so I would have liked the >> Yeah. Yeah. So we can see that um the Seahawks um before the wild card round they were at 27% to win Super Bowl and

8:15they're now dropped to 25% because their bracket has gotten harder because the good teams won last week which is very sad.

8:23 >> Yeah. Um but true and we can see which teams improve improve their odds the most. The Texans did uh along with the Bills uh Rams up there pretty pretty nicely as well.

8:34 >> Yeah. >> Which I think all kind of >> insights too which is great.

8:39 >> Yeah. Yeah. So it's giving us some insights. It tells us about this new view. No changes to these ones. Okay. So um I really want to see if it can drill into why these what caused these percent changes to Super Bowl odds. So, I'm just going to ask what caused these percent

8:56changes to the Super Bowl odds.

9:00Look at each of the six games

9:05uh to allocate uh this change.

9:09Great. So, um let's see if it done this before like if you didn't weren't using MCP and Claude like and you were trying to just answer this yourself like what would you have had to do?

9:21Yeah, it's a really good question. You know, I would have um I think I think what I would have done is uh wrote these SQL queries myself, right? Um without the MCP, I probably would have done a little bit of AI AI assistance, but assuming I didn't have AI at all, you know, I would um I would I would do very

9:37similar analysis to this, but basically what you would what I would do is look at um basically in I have an underlying table that has the results for every for all 10,000 simulations. And then I would just basically look for each combination for each team which which kind of um pairs impacted each other the most.

9:55 >> Um it's it's actually fairly complex SQL to do. >> Seems like >> um and well SQL is like actually SQL is actually kind of hard to use for data analysis like zooming out. Yeah.

10:06 >> Um which and the reason is it's very verbose, right? um you have to write a lot of words to do simple transformations, especially if you compare to something like pandas, >> right?

10:18 >> Um but on the other hand, we have 50 years of training data for SQL. So if I have an AI doing it, I don't really care as much that it's using a lot of words.

10:26 >> Yeah. >> Um >> continues to fix itself. Like if it writes it, it's like, "Oh, I hit a wall." Then it just continues to work through those errors.

10:34 >> So actually what's interesting is it's doing very much what I would do, which is like iterate through this.

10:38 >> Yeah. Um uh and it was what's helpful is it actually puts it what it's thinking as a comment in the query.

10:45 >> Yeah. >> Right. Um and so you'll see it's it's it's writing a CTE here. Okay. Where the winning team was these two teams like how does this how does this game right this there was a specific game where these two teams could have won in this in the simulation. How do changes to that impact the Denver Super Bowl odds?

11:01That's the specific type of query it's writing. Now now this is very very specific. It's probably more specific than what I would do. Um, but we can see

11:12we can see the Super Bowl odds with the Steelers were high. If the Steelers won, their Super Bowl odds were 23% and if the Texans won, their Super Bowl odds are 20%.

11:20 >> Right, which makes sense. >> Um, so like it's doing it's doing these very atomically. My assumption is the next one is actually the Seahawks query.

11:27Let's see. Uh, no, this is actually someone else. Uh, looking at New England versus the Chargers. Okay. I thought it was going to look at So, it's still exploring this data. It's doing unions.

11:36It's writing the information. Um,

11:42very cool. Okay. So, we can build an attribution analysis. Uh, here you can see it's it's using cat here, which is very fun to uh take this uh table um and

11:53make it so it can consume it here. So, um yeah. So, here's a breakdown of how this game impacted the odds, right? Um,

12:03so see, so the Pit, so Pittsburgh lost, which means their Super Bowl odds went back to the pool, right?

12:09 >> Um, and so they lost. [laughter] >> Yeah. Went back to the pool. And so now we need to reallocate that to someone else.

12:17 >> Yeah. Yeah. >> Right. >> Um, uh, this is this is an interesting statement. uh the margin blowout may have boosted yellow confidence if I had implemented that feature of my model then it would have it did not so >> um you know something to note here I'm obviously built this model so I kind of know my way backwards and forwards in it

12:36 >> um and again it's now so it's now doing this kind of tented analysis for each each one of these um this game didn't really impact Denver Jacksonville and Buffalo are not you know particularly great teams as in terms of their ELO rating this year okay um so we got this analysis over and over. Where's our Seahawks? Okay.

12:56Seahawks drops from 26.9 to uh

13:01Seahawks slight reverse facing SF over Philly. Yeah, that's right. That's right. We would >> we would do that.

13:07 >> Um >> this one is this is we our schedule got harder. >> We dropped by almost 7% by if Carolina had won to if the Rams >> I know. I know. We like we're ready to go to the Super Bowl, baby. Um come on.

13:22Let's so uh this all looks pretty good. It matches what I what I would expect.

13:27Um okay, great. So we have an analysis here. Um let's like formalize this into something um build this into something a bit more formal using invis.

13:40 >> Okay, so this is telling it to use um to use my skill, my charting skill. So it's going to load in, look at the documentation, see how you and then [snorts] it's going to create something in cloud called an artifact. Um, an artifact is just a flat file. Um, in

13:57this case it is HTML. Um, I think you can build JavaScript apps. I think you can maybe even build like full-blown apps with um uh with this. So it's I guess artifacts can be more than flat files, but in this case I am using um uh I'm using these artifacts to generate HTML. So in this case, well, I'm using a markdown

14:17artifact to generate an HTML artifact. So it's giving us this markdown file here. Um, my skill basically says here's a way to use markdown and JSON to define data visualizations and commentary and so we can use that skill then to uh make HTML. It's actually an interesting kind of story around how we got here. Um, as Logan can attest, we've been doing a lot

14:36of data viz. Um, data viz is pretty expensive for cloud and um, if you've done much of it, you've probably crashed it at least once. Is that true, Logan?

14:46 >> Yeah, absolutely. Maybe once a day at least. [laughter] uh once a day after learning how to crash it. Much less than once a day [laughter] probably.

14:55 >> Yeah. All right. So, it's generating the HTML, which is basically running a little Python uh inside of the user space that's available here, and it will generate uh HTML for us. Okay. So, let me widen this slightly. Um I don't actually love this chart. I don't love this spark line. I'm not quite sure what it's trying to tell us here. So, we're

15:16going to do a little bit of riffing on it. Um here's the lost equity. How much team how much uh how much kind of Super Bowl odds were added to this um by based on that. Yeah. Gave the most back to the pool. Jacksonville and Philly.

15:30 >> Yeah, those two teams were like, you know, pretty good teams for ELO. Um I don't love this analysis.

15:38Oops. Okay, hit the back button. Let me hit the forward button. All right, let's um I don't love it, but that's okay. We'll modify it. What do we got here? Uh winner pre winter post.

15:53Okay, impact to the team. This makes sense. This is a nice little table. Here we go. Here's our Seahawks impact. Um

16:03okay, so we can see we have uh

16:08some changes to to these probabilities as well, which is very nice. Um, all right. I'm going to do one more thing here, which is can we make a matrix table that shows us

16:24actually I'm going to fix I'm going to fix one thing. I'm going to fix one thing that's much a little bit simpler so we can hop into something else. This is kind of the basic the basic flow here is now we can start iterating on these like these trends. I don't I'm not sure what it's trying to tell us here. We

16:36probably need to be more specific in our prompting there. So, I'll do that. But what I really want to do is just make a small change, which is the pregame win chart.

16:45Make it the same height as the up upset impact summary table.

16:55Um, and so we have little controls kind of in our um uh in our uh dashboard that let us make very small changes to it and then regenerate it really quickly.

17:09And we'll see that this chart will now get a little bit higher. Actually, that didn't work. Went from five to four.

17:15Let's change it to change the height

17:19to eight. It should work. Find out.

17:24So, these are kind of like little things we can do to tune this. Um, we can also see our percentages aren't aren't formatted correctly here, too. So, some small issues, but that's okay. Um, those are very easy prompts to fix. Um, I will not deal, you know, jump too much into the minutia here. And hopefully this fixes my height of this pregame

17:44probability table. >> Yeah. And then Oh, there we go. >> Oh, perfect. >> Awesome. All right. So, we have a little bit of an NFL analysis here and it tells us something interesting. So, >> how are you feeling, Logan? You ready to you ready to hop in?

18:00 >> I love that. Yeah, I think that's great. I mean, I think the thing that's cool about that, too, is like I there's so much you could continue to ask the model and play around with. I'm like, who would we most want to face in the Super Bowl? You know, who would we least want to face in the Super Bowl? You know, if

18:15you like high scoring, what would you go after? Things like that. Like questions you could just ask of the data pretty easily. So, that's the the fun thing.

18:23 >> Totally. Totally. And you get a little some some charts on it. Um, all right.

18:27I'm gonna stop sharing. >> Okay, let's do it. Okay. >> So, >> you're live.

18:34 >> I'm live. Okay. So, here's what was interesting to me. We talked about this the other day is on the golf side. If you've ever tried to look at like the PGA Tour uh data, and I I brought my polo, so if I need to like dress up in my nice polo shirt for golf data, let me

18:50know. Um, but uh it's not very easy to

18:55work with. Uh it's like one category at a time. It's like the top three and then you can click detail. It's not great. So I wanted to figure out how I could like pull in more data to actually start to ask some interesting questions about it.

19:09So pulled in that data last night and wanted to start seeing things like okay let's inspect the data. So loaded to motherdoc and you know starts to break out some basic things just kind of hey here's some of the data we're seeing in the database. Um, you know, here's available stat categories. You know, Scotty Shiffler was dominant. Not not a

19:30big surprise. Um, one thing I wanted to see was like correlation to money list.

19:36Um, specifically in stats we might be able to track. So like which categories would matter most based on what we're seeing so that we could start to analyze, okay, how did somebody do relative to that category and and could they see improvement in the future?

19:53Things like that. So obviously like consistency wins, things like that. But some interesting things stand out to me like shots gained te to green vastly outweighs say shots gain putting, right?

20:04So like work on your putting. Yeah, of course. Uh but like really make sure that like te to green you're a spectacular player because otherwise like the putting piece is slightly overrated. You can make a lot of money by being a decent putter. You don't have to be the best.

20:20 >> Interesting. So that kind of stuff to me is like super interesting. Um, and I wanted to kind of correlate like does that hold. So the data I found was last year is pretty easy and then also pull the data set from 2017 to 2022 and put

20:35that in mother duck as well. And I just wanted to see like do those hold relatively constantly like over time.

20:41And so as an example you could see like 2022 to 2025 like pretty consistent.

20:48There's some minimal changes, but like for the most part, these tend to hold pretty true. Um, and so for me, like if I'm honing in like obviously top 10 finishes, like that that makes sense, but things like shots gained or shots gained te to green, like you know, that gets more interesting. So like this >> the thing I want to explore is like okay

21:08 >> for shots gained and shots gained T2

21:15green who uh played well in 2025 but

21:21perhaps underperformed on the money list because what I'm looking for now is like >> like help me understand who like actually statistically like did really well last year, but maybe there were just a couple of like big first tournaments, majors, things like that where like that's not where they played quite as well, but maybe this year could

21:47be the year they would pop, right? Like, yeah, >> over the course of a whole season, they do these things really well.

21:53 >> Um, and therefore, if I'm, you know, just that person, I'm excited. Like, cool. If I keep it up, it should show.

22:01 >> Sure. But also if I'm in any sort of like you know golf pools things like that like oh like look at this this is great. So to me this is super interesting right? So like okay so Pearson Cudi Pureborn Olison you know couple of examples here like their earnings okay I would love to earn a million dollars playing golf. Uh but in

22:21terms of their ranking that had them like 124th or 107th like not even in the top 100 on the tour. Um, >> and but they were actually like significantly better at what we see are like key correlative like categories, right? Like so shots gained their rank was 28th >> or 20th here um in the case of those two

22:43guys. And so like to me that gets really interesting. It's like these would be a couple of names that I'm you know going to be watching this year like oh cool like that's really interesting. I would think this year would probably be a year to keep an eye on both of those folks is like pretty likely either of those could

23:00pop as like a Darkhorse person. >> Yeah. Yeah. Yeah. That's super interesting. I This is great. And remind me shots gained is like basically a statistical calculation based on like how far away you are from the pin >> and like how close you get like post after the shot. You know, you rate a shot based on are you closer or further

23:17away or something, right? >> Correct. So shots gained would be like compared to your peers in that specific category, how many shots better are you?

23:26Right. So like, okay, a pretty fascinating thing is like uh because we have the data from 2017 to 2022 and last year like Scotty Sheffler last year amazing year. So like how did Sheffler's

23:402025 in terms of shots gained in key

23:46categories compared to the best players in the

23:53prior years data we have. Right? Because this >> this is one of those things that like >> you watch something and you're like, "Oh, that seemed really good." or you hear talk about it and they're like, "Oh, that was the best we've seen." You know, but it's like, is there a recency bias in that? Like, is it really the

24:11best we've seen? Like, that's some of what I love to take and really understand with data like this is like >> 2017 to 2022. It's like a pretty good stretch of time, like fiveyear period.

24:23And like this was both a available data set um that I could grab pretty easily, but the other is like this is pre-live.

24:31So like before the live tour shows up and like starts pulling away some folks like this is you know a fully stacked you know >> minus anyone new that's joined the tour in the last couple years. This is a very robust PGA tour right so like if you're in the top in some of those categories during those years like you are a

24:52phenomenal golfer. Um but look at this like by far the best season in the data set right? So this is actually

25:02important, right? We could see that in terms of his category versus whoever was the previous best in any other prior year like probably the best year in this prior data set is like Rory's 2019 Sheffller's better than him in total shots gained. Uh TD Green shots gained shots gained on the approach. He's better than Morava in 21. So like he is

25:24statistically better in all the categories that we think align to you know money earnings compared to all those prior years. So it's kind of it's an historic dominance >> and so anyone who speaks of his last year in that way like is accurate like that.

25:40 >> I mean I don't think we need the data to show us that but is [laughter] very interesting to put it in context.

25:44 >> Yeah for sure. And uh the only other thing like I'd be curious about is like uh yeah I like this. This isn't just the best season of your day. It might be the best ball striking season ever. You know, it's like [laughter] that's probably true. I don't have the data, but that's that is how it felt. Um, one

26:01other question though that like I was thinking about is like um can you look

26:07at averages from those seasons?

26:13uh to see if he's only better because

26:19the averages are worse, right? >> I see what you're saying. Like, have the other golfers gotten have the other golfers gotten uh worse?

26:29 >> Yeah, because let's say that you are Rory in 2019 and you're like wanting defend to defend your year like he wouldn't do this. He's like an anti-live guy. But like if he wanted to, he could say like, "Oh yeah, but back then Dashambo, Kepka, all these guys, they were still around. Like they hadn't gone to live yet." Um, and meanwhile, like

26:49now they're coming back in their, you know, in their shorts and hands saying, "I would do want to play 72ole tournaments." Um, for a few years that wasn't the case. And so, uh, could he

27:00defend it and say like, "No, no, no, no, like it's easier now." or does that part not actually even matter? Right? Like actually is just the the data showing that his shller's 25 was better no matter what. Right? So that's something I'd be curious to to play around with here. Um and we'll see if it's able to

27:20put something out here in the next minute. But that that's the kind of stuff that I get interested in. And then like you know longtail like playing with this data you know after this chat like the other thing that I want to figure out is and it's pretty easy to do this is like the majors this year like one of

27:36them's always at Augusta right the masters so that course doesn't change >> but the other courses that do change like playing well there how much does that matter um what categories matter most and who did well last year in those categories and can I pick like a front runner and a dark horse based on 25 for each maj [laughter]

27:55pool pool. >> Yeah. And just put it in my golf pool, right? Yeah. So, okay. So, this is interesting. Like, you're right. The 25 field is weaker. The top 10 average dropped from 1617 to 39. 10 best player would barely crack the top 20. Here, that makes sense.

28:11 >> But still, good players going to live. Right. Right. Right. >> What still stands out? His gap to number two. Kind of a a zcore standard deviation above the mean. Even against a weaker field, separation still unprecedented. So, >> yeah. Yeah, you know, >> pretty fun to analyze and see that way.

28:25And we could use MVIS and all sorts of things. But >> you want to um uh I have a couple more minutes here, Logan. Why don't you pull up one of the analysises that you posted on LinkedIn yesterday? I thought it was just a really awesome way to kind of take a look at this.

28:37 >> Yeah. Yeah. Yeah. Uh let me see what >> here. Let me here. Let me I'll hide you for a second so you can fish through.

28:43 >> Okay, cool. >> There you go. Now you can fish through and find it. Um yeah. Yeah. This is um

28:50Yeah. One thing we haven't really talked about too much is there's a bunch of ways that you can visualize this data, not just using like the skill that I was showing off earlier, but also um there's full-on React and front-end skills inside of Claude. So, you can make some really really cool stuff.

29:06 >> Yeah. um uh you know with fairly minimal

29:11um fairly minimal prompting um >> obviously kind of uh what it looks like longer term to like build something reproducible is a little more challenging you have it pulled up pulling it up here we go >> sir >> yeah so there's a few different ways in which I was trying to again analyze that same type of data from let's say last

29:32year and you know what do you look at and like one example is like Okay, let's, you know, do something like this.

29:38This is for me like pretty fun, right? Because like it's one thing to just see the data. This, okay, let's go shots, game, total. We just learned that this is like one of the most correlative to like how well you'll play. And you see a bunch of data and you're like, oh, a gap. When you actually see it on a line,

29:54okay, the green are the top five. Look how far ahead. >> Wow. >> The second dot.

30:01 >> Same with the green. It's kind of insane, right? Like that's uh it just really gives you this like visual understanding of like in the most important areas like how much he was dominant in that way. Um, a and then also down here was kind of interesting like I also wanted to look at just hey who's top 10 in different categories and

30:22like oh he had four top 10s some of these people had three but like he was number one in three of the four times he was in a top 10 and number two in the other category like >> you know it's like really visually you get a sense of like oh okay it wasn't even close right so yeah that's just

30:41some of the visuals that you can build I think there was a couple of other examples that I was playing around with and built out. Uh what's this one would we >> Oh yeah, this was another example of like utilize this data um and get a sense of again like you know same type of data but how are you looking and like

30:58hey PGA tour if you're listening if you're one of the people like make this available like this should be how people interact with the data this is like way more interesting. Yeah, give us the claw app. Where's the Where's the Claude app on the PGA Tour website?

31:11 >> Yeah, that would be sick. >> Yeah, let's around the corner and do this. So, >> that is great. Logan, thank you so much for hopping on here and sharing a little bit of your expertise. Um, >> uh, this is this is phenomenal. Um, thank you everybody for joining us. Uh, we'll be sending some follow-ups via email. Um uh I will uh uh well it looks

31:36I've got a question to take take an ask on real quick. Um we'll we'll take this question and then we will wrap this up.

31:42Um okay the question is how do you make the viz this nice? Um Logan do you want to answer?

31:49 >> Well you helped me build a really nice skill and so using the invis skill is like a good example of like just referring to that which is great.

31:57 >> Yeah invis skill is great. You can also uh all of the kind of people who are well-known academics and write about um data visualization like tuy for example you can just be like you know hey build a tuy inspired chart for this and it does a pretty good job of interpreting that um without too much input from your

32:15side. There's also like I said a front-end skill that can use react and so you can really tune these to make them look really tight.

32:21 >> I also think the fun thing with this stuff is like it's somewhat subjective right like you should pick what you like. So like before I even talked to Jacob about this, I just said what are five different like you know academics and the visualization or like types of visualizations and take this data and represent it all five ways and then I

32:37just looked at each and I just picked what I liked most. >> It's about taste. That's right.

32:41 >> All about taste. So like that's another option. It's like make it your own.

32:45 >> Yeah, totally. Love that. Logan, thank you so much. Um, we're really excited to uh continue these MCP sessions and talk about what we can do with our friend Claude and the mother duck MCP. Thank you everybody for joining us. Logan, thank you. And we will catch you all next time.

33:02 >> Go Hawks. Keep it in the short grass. >> All right.