Turn Claude into your marketing analyst in less than 30 minutes with MCP

0:00This might be one of my favorite VI marketing use cases yet. This MCP integration with Claude allows you to visualize and interact with your Google Analytics data. No one likes spending time in Google Analytics, so check this out. I think it's going to make your life a whole lot easier. Enjoy. All right, so been digging deep into MCPs for vibe marketing purposes. And today I have Veru here from our team. She's our amazing automation specialist, growth manager, helps with the community, helps build a bunch of awesome workflows uh with me. Uh so imagine you're working on a team and your boss is hassling you to understand how much traffic you got last

0:44week and everyone knows that Google Analytics 4 is a pain in the butt to log into and navigate. uh you know, let's just figure out how to chat with a with an AI model to get insights from from our Google Analytics so that everyone can save time. And then maybe there's other cool things you can do if that data is living in cloud. Claude can build things. It can format things for you. So maybe you just want to put reporting on autopilot in a sense. So

1:11let's have a look at how this is done. I'm going to pass the mic over to Veru and I think this is going to blow your mind. On to you. Thanks James. Thanks James. Thanks for the intro. And I think that's that's a pain point for me as well like you know going to Google

1:25Analytics 4 and looking for the data. It's very difficult. So I probably bug you with a lot of uh a lot of questions in Slack for numbers and all that. So hopefully this will uh this will help save you a bunch of time as well. I hope so too. Let me just share my screen. You are able to see my screen

1:47now? Yep. Okay. Um, before we get into this whole workflow, do you want to give a quick intro of what MCP in general is and then we can go here. Uh, what do you think? I think just kind of go through it. Like I think MCP can be a very complicated topic for non engineers and I'm not an engineer and I probably wouldn't do the best job describing it. my from my basic understanding it allows you to uh interact with you know data and things within your you know clawed uh local desktop instance. Is that is that a basic uh understanding? Yeah, I think to to me uh I've understood it this way. I

2:30like you said it's it's very hard for non-technical or non-engineer to understand. So basically anything that you have in your desktop uh you want to basically access that or you have something that is stored in different tool it can be slack air table anything and there are so many things to access. So how do we basically chat and then access it's as simple as you know your virtual assistant by connecting chatting with multiple tools and getting you what you need to uh get it done. So I think that's something a simpler definition here. Yep. Sounds great. Let's see how it's done. Yep. So with MCP server trigger, uh if you see

3:10over here, right? So I if you there's like multiple nodes over here. So this is something that I learned that when you have MCP server trigger, it's like best to have multiple tools associated even though this is G Google Analytics 4. So this is like mult same Google Analytics 4 but with different functionalities. For example, this will

3:32be looking into the traffic sources. This will be looking into the visitors data. This will be looking into the engagement data. This will be looking into the uh you know top performing pages and things like that. So when why we use this is because it's a very simple thing. Um MCP server cannot handle all the commands at once. Like you know you have to send a command based on the command MCP actually runs

3:57on an AI model. Here we've used claude. So it picks the data. It runs based on that and then it basically pushes out the output. Yep. So I'm I'm going to get into this a little bit how we do this. Um right now when we do Google Analytics for connect uh there needs to be like multiple things. One is Google Cloud Console. Uh let me just quickly open but I think I yeah I think I'll just let them know. I mean like uh there's like a video on YouTube that they can see or use chat GPT here where they have to go to Google cloud console and in the Google cloud

4:35console they create a project and they have to create an API for oath uh 2 specifically for Google Analytics data. Okay. Um then you know basically there's like client detail, client server, client ID, everything they get from there and they also have to give access to scope like whether it want to read, edit, delete all the uh you know analytics data. Okay. And then this is set. Okay. So uh you have a a node here where you are uh connecting to the Google Analytics data which is being piped into Google Cloud Console. So, Google Cloud Console is storing the data in a way that can interact with, you

5:18know, the chat model and and things of that nature. The way that you're getting that data into Cloud Console is through this OOTH setup step where you're probably going to have your, you know, property ID or something like that for, you know, the Google Analytics account that you're wanting to do this for. And then you will give it certain access permissions. We don't want it to delete the data or whatever. So uh I'm assuming that you just use something like read or or something at this point. Is that right? That's perfectly right. We only give readonly access. And in fact in the Google cloud console uh you will be able

5:53to see that hey this is restrictive scope. This is uh like a very secretive scope. So I would suggest you not to use that. So we can always remove it. Uh and read only is the best option. Okay. Makes sense. Cool. And I think what we'll do is put like a like a one pager uh under the video in the replies too just so that you know people can understand exactly the steps to go through uh to set this up. Yeah. And the only reason we're not showing it is that we have all the secret information over there. Yeah, we don't want to we don't

6:23want to share that. Yep. Sounds good. Awesome. So the next step that we need to do over here is very simple. Uh I basically use chat GPT here to get this JSON. Um what we are going to do is let's say consider each HTTP request node as a tool. So this is a tool that's going to fetch data uh calculate data or any data around uh users and user engagement duration. Um so that's what this is going to do. So that's the JSON that we are adding over here. Y similarly if you see over here again this is going to be like the session source uh screen page views

7:03engagement rate. So this is a tool that's going to look into all the engagement rate related data. And if you specifically want to get bounce rates and stuff like that you can also like use this. Cool. So basically if you create these sort of custom uh JSON instructions or whatever you can you know manipulate the data in the sense of coming up with custom calculations and making sure the model knows exactly uh what to access there from the data set that you're providing it. Is that is that fair? Exactly. So basically we are uh MCP uh server trigger command and this is basically taking us to which

7:44where exactly the data has to be like uh fetched from and there are these tools which is let's say traffic tool top page top performing pages tool so each does a different functionality over here. Yep. Okay cool. So what's after uh this step right? So much before that when we connect this MCP server trigger we are supposed to uh download cloud for desktop y uh and there is an option in cloud for desktop. I'm just going to open that where you have to go to settings and there is something called developer and you have to like configure it in your terminal. So this has to happen within your uh desktop. Okay, so

8:29this is completely a different step and if you want a documentation I think we can possibly share the link to the documentation that cloud has shared. So that step should be really very helpful. Okay, cool. Yep, we'll share that as a follow-up too. So you have cloud desktop set up, you have this kind of workflow built out in in so uh what comes after

8:47that? Yep. So everything is set up now. The command server is set up. The tools are set up. Now we have to chat with Claude to basically retrieve the data. So the data has been fetched. We are going to retrieve the data. So this is the fun this is the fun part. This is the fun part. Right. So let me say something like So I see your inade screen right now. Uh just switch the uh

9:11screen over to Claude. Just a moment. Yep. I'm just going to switch

9:24Are you able to see now? Yep. Perfect. Awesome. So, I'm going to ask Claude to basically get me share average session duration per month. Okay. So, it has all the data and possibly you can see it over here that this is connected. Yep. Yep. Wow. So, it is also creating a visualization for us to like look at it. That's incredible. So, you need to make a report, you need to make a PowerPoint, you need to do whatever to uh, you know, give your clients or your boss some something and uh, you know, you can just come here and literally ask

10:09for claude to create your visualization. That is insane. That would save people so much time. It's crazy. Exactly. And it also gives key insights that when which month was actually the longest session time and you can always go back and check why it was the case. Wow. Okay. So the next thing you could probably do if there's a certain metric like that's off, you know, like just looking at this table. All right. My session duration in August was down uh 20% or whatever. uh can you say like give me ideas or hypotheses or something that I can do to improve my session duration you know so I think what's

10:48really cool about getting this data into claude is that you know you create these loops these that are that are quick so uh you know claude could probably give you some ideas right here to go and improve the metric that you're talking about or if your boss is like oh what's happening here like how do we get that up it's like hey I I'll let you know in like 3 seconds I'm just going to like

11:10prompt claude and get some cool ideas. So yeah, uh the the speed at which you can a get the insight and b put it into action uh just drastically compresses here. Exactly. And you know what I tried to do something interesting as well like for example I've I've run bunch of different sessions here and I've asked you to create like an entire dashboard like for example we've connected different tools to MCP server. Can you create a dashboard with all the

11:43data HTTP node?

11:51Okay. So, this is looking at all of the uh different nodes that you've connected. So, it's taking into account the visitors, the session information, and things like that. Okay. Exactly. That's pretty cool that you know you can set that up separately to keep it you know clean and make sure the data is accessed in a usable way and then this is like automatically kind of like blending that that data together uh I

12:16guess which is which is pretty nice. Yep. And you know you can also like uh prompt it so that you know you can add additional sections in the dashboard and this dashboard can be real time and things like that and you can change the color because it says that this is the color it's using the brand color that you want. You can do all that here in

12:35the prompt. Wow. Yeah, this is uh this is doing a lot more than I thought it would. I thought you were gonna show me like a table that had, you know, monthly visitors over the last six months or something like that. And, you know, this is another like uh, you know, I've talked a couple times on X about like uh, you know, marketers within like companies and things like that have an opportunity to just like show 10x value to their teams. You know, if you're the person that knows how to get the data here, interact with it, uh do that easily, get insights quickly, create visualizations in a snap, I mean,

13:14you're like a 10x marketer uh on your team, which is just amazing. And it doesn't even take that long to set something like this up. I agree completely. And I myself did not know that I was able to like you exactly said but I just tried like what if I can get a graph out of it and then you know the next thing is like magic I was able to like get of that. Yeah. Yeah. Some reason there is an issue but I think no issues I have already created a dashboard before so I can show that you

13:48are always prepared. Yeah. Awesome. Look at that. All right. I saw our traffic going down a little bit up there. Uh, we'll just say it was seasonal. I'm just kidding. Wow. Yeah. So, you all of it. This is awesome. This is awesome. Very cool. All right. Well, um I think people are going to love this. And uh yeah, let's give them some step

14:13by steps for how to how to set this up. And uh I don't know, this is just a great vibe marketing use case uh right here. I think it's awesome. And uh well done, Viru. Thank you. Did I pass the vibe check then? So you're going to ask me any uh data I'm going to like chat with Claude and get it to you. Pass the vibe check. I mean all you have to do is put in a prompt and all the boring manual stuff's happening in the background and you know uh that frees

14:41you up to to do other stuff. Looking forward to that, James. Cool. Awesome. Well, thanks a lot, Veru. Great work and uh hope everyone enjoys it. Buffy.