TEA 241: AI Can Now Run Your Meta Ads For You (Should You Let It?)

Introduction to MCP

0:00Meta just dropped something that 99% of media buyers don't fully understand. It's called an MCP, and an MCP is going to change the way that pretty much every direct-to-consumer brand interacts with their ad account. In this episode, we're going to cover what an MCP is, because some of you listening are probably like, I'm already lost, what Meta's new MCP can do, and what it means for you, your team, and your competition. I'm really excited about this. This is the Meta jumping in on the wave of AI, but like I said, I want to start talking

0:35about what an MCP is, because I think that is ground-level AI understanding to really start to utilize AI in your business.

Personal Experience with AI

0:46Before we actually get in, hey, welcome back from Tokyo. Oh, thank you. Yes. Yeah, I was just in Tokyo, Japan for two weeks. We went to Disneyland, and then we traversed the city. I actually sent an email out through Breezeway, and it was like, I felt like I had to use Google Translate for everything, because all the menus were in Japanese, all the signs, everything

1:17was in a lot of it. There was like 50-50, and I was like, it kind of felt like Google Translate was like my bridge between understanding and something I didn't understand, and that's how I kind of felt like Breezeway was, because Breezeway feels like the bridge, or the Breezeway, which is why it's named that. The bridge between meta-ads, not understanding meta-ads, and actually understanding meta-ads, Breezeway helps with that. So I sent out an email. So if you guys were on the Breezeway list, you probably saw that email. That was basically my Tokyo recap in an email.

1:50There you go. Heck yeah, dude. Just another interesting little thing I would say, I know you're mentioning MCPs and stuff like this, but this is done through the connectors in Claude, right? Yes, yes. But it's not just Claude. This is like open to others.

What is MCP

2:04Yeah.

What is MCP

2:04So what I'm going to talk about, what an MCP is. So I guess that's a good segue. So an MCP, it's called, MCP is shortened from model context protocol, model as in like an AI model, large language model. And the way you can think about it, I've been using MCPs. I think you've used some MCPs as well, correct? MCPs are essentially the new version of an API. So if you know what an API is, or you've heard of it or used it, this is like the AI version

2:34of an API that essentially allows the large language models, Claude, ChatGPT, anything that allows for MCP connection, which I think pretty much everything does now. I know I only use Claude. So that's really the only one that I know in particular, but I think chat. Even Higgsfield, which is an AI creative gen tool, go to Higgsfield.ai if you want to get into AI creative. They just integrated with Claude natively, I believe. Through an MCP. Through MCP.

MCP Connection and Usage

3:01Yeah, so MCP is essentially the way that software and other technologies connect to the AI. And some of them have like actual native connectors. So if you're in Claude, you probably have natively connected Google Drive or Gmail or Calendar. Those are all like native connections. But for some MCPs, such as the meta one, you have to set up a custom MCP, which really isn't that hard. Meta gives you a URL. You take that URL, you put it into the custom connector, and it's done.

3:32It's not that hard. Now, there are some that are a little bit trickier. Like we use GoHighLevel for our business. And the GoHighLevel MCP, I got it to work. I don't know if you've played with it or not. No, I just heard how terrible it is. It's awful. So the GoHighLevel MCP has to run on your computer. You have to have a server on your computer running it. And then it sucked anyways. It was really bad. But this is an easy MCP to install, fortunately, for all of us. Now, here's what's really interesting about MCPs. They allow AI to read data from your tool.

4:05So it can look at anything that the MCP gives access to. Sometimes MCPs are a little limited, like an API. But it gives access for it to read from the tool. But sometimes it can also give access to write, which in meta case would be creating a campaign, an ad or an ad set or anything. Taking actions. Taking actions. Inside the tool. Yeah. So and the other thing that's kind of cool is an MCP gives the AI context on how to use the MCP. So if you've ever done something where you're like, hey, I'm going to upload a spreadsheet and I want you to do something with this spreadsheet, you have to tell the AI, you know, there's this column here and this column here.

4:45This column relates to this column. And I want you to look at this column and determine, you know, you have to give it all of that context where typically, depending on how the MCP is created, the MCP already kind of knows how everything works. So when you say, hey, go get the data from this ad account or from this day or whatever, it knows just how to go and get that. You don't have to give it as much context as you probably would if you weren't using an MCP. So that's like the baseline of MCP as a whole, how it actually works in the AI ecosystem.

Meta's MCP Capabilities

5:16But let's move on to what Meta's MCP actually does. I just want to say that like this is, well, two things. Number one, this is really groundbreaking MCPs because API has been the longest thing that I could ever remember. API has always been it. And this is like, so the fact that there's now a shift in something else from API. API, it was like API integrated. Does API integrate with this and this and this and this? And then now MCP is becoming a thing. It's crazy. Like the world hasn't changed from API in decades.

5:47So the fact that this is happening, I think everybody needs to take note. Like this is really, really big. Something like this doesn't happen and then become massively adopted unless there's a big shift coming. And it's coming. It is very, it's coming very quickly. And it's very exciting. I'm so excited about this time. Also, I think this, our thought on and conspiracy theory on this is that I think China blocked Manus from being able to be purchased by Meta. I didn't do any research on that. You didn't? No. I think that because China blocked it from the acquisition of Manus from Meta, Meta's like, okay, fine.

6:19And you know what? Because previously it was like Manus was going to be the one. It was going to be the one. It was the chosen one. You're the chosen one. Yeah. And what it felt like. Yeah. And it's like everyone is talking about it too. And then people are using Claude to do stuff and their ad accounts are getting disabled and stuff like that. And we hear, we heard stories and stories and stories of people doing that. So everyone's like kind of freaking out and scared about it. And then suddenly China blocks Manus from being, I guess, sold to Meta. I don't have all the super details on it, but I know they get blocked. And I'm talking to our contact in Meta about it. And they told us like, yeah, it got blocked.

6:51And that's probably why they decided to just be like, you know what? It got blocked. Here you go. Everybody. Which is wild. Because if you think about it, like Meta was relying on that sale to go through to charge you to use AI. Brilliant. Because they wanted, like you would have had to pay for Manus Max or whatever their thing was to use the AI to its fullest capabilities. And that money would go to Meta. But now that that's blocked, if that's the case, which like I said, I didn't do any research on it.

7:22But now that that's blocked, you just use the free MCP through whatever tool you're already using. Now, we know Meta is working on their own LLM. So maybe that will create something. But it'll be interesting to see how that even goes. I feel like Manus was probably going to be the backbone for that. So now that that isn't going through, it'll be interesting to see what happens to Meta's LLM. So here's what you can do with the Meta MCP, or at least what Meta says you can do with the MCP.

7:53It can do reporting. So you can pull ad account performance and the data from your ad account, which, you know, a lot of people were already just exporting CSVs and uploading that to Claude. But now it just does it natively. Ad creation and management. This is kind of cool. You can build, launch, and edit ads through the AI. Catalog creation and management. So you can manage product catalogs for shopping ads. Signal diagnostics. So you can analyze pixel conversion data and the quality of your pixel, which I think that could be interesting. Because I think a lot of people, the pixel is like a black box of, like, I don't know what's actually going on in there.

8:27So now you'll be able to ask Claude or whatever AI you're using, what's happening with my pixel? And it might even be able to help you diagnose. Because a lot of the times for our clients, they have a pixel issue. And it really is a black box, typically. You need to have a lot of understanding of how pixels work to actually fix a pixel. So hopefully now with Claude, you'll be able to say, hey, my pixel is not working. Or is my pixel working and what's your analysis? Yes, yes. Here you go. And the other thing is it can do help in troubleshooting. So it will actually, like, research meta knowledge base and give you recommendations based on that.

9:00So that's what meta says the MCP can do. Sounds really promising. So I was super excited. I wrote this nice prompt. And I wanted to have it analyze the eCommerce Alley ad account. So I said, hey, use the Meta Ads MCP connector to take a look at the eCommerce Alley ad account. Specifically, take a look at the LTO performance campaign day over day for the last 14 days. So I told it the specific campaign that I wanted it to look at. I'd like you to act as a professional media buyer when analyzing the campaign. First, ask me any questions that you need to know about the campaign.

9:32Then dig through it. Structure, settings, creatives, links, and data to find any holes. And then I gave it some potential holes that it could find. Creative types that we haven't tested. Copy relating to poor performance. Pixel issues. CPM issues. Or anything else that you think may be negatively impacting the campaign. And then provide me with a visual report of your findings. So I typed out that prompt. I was really excited just to see what it would spit out. And the first thing that happened, which was kind of, I almost expected this.

10:03I told it the name of the ad account. The eCommerce Alley ad account. And then it couldn't find the ad account. And I was like, okay, I kind of expected this. So I went and I gave it the ad account ID. Because sometimes even for us, like when we're coaching clients, we sometimes even when we go and search the ad account name, it doesn't show up. So I was like kind of unsurprised. So I grabbed the ad account ID and gave it the ad account ID. It was like, hey, guess what? I found the ad account. I was like, okay. First hiccup, done and over with. And then hiccup number two was after we got it, it returned an error that said MCP is rolling out slowly.

10:36And it's not available to your account yet. Awesome. I was like, great. So this is super, super valuable. And so I then went back and looked at the meta post that they did about it. And at the very top, it says, you may not have access to this feature yet. It doesn't specifically say, you know, how many people have access to it. But I did some deeper research that I'll share here in a second that very few people have access to it. So even though there is technically an MCP, you probably can't use it.

11:07That's my assumption. Most people probably listening to this probably don't have access, but it will probably be available within the next few months. So my first impression was pretty much, I couldn't have an impression because it didn't work. But I did do some research on X to see if other people were having, if they were able to get it to work and what they were able to get it to do. So in my research, it was actually kind of disappointing from what I was finding. Some people were reporting that metrics weren't even accessible, even though they are in the API.

11:41Because meta does still have an API. That's what Breezeway uses to connect to meta. And even though it's in the API, for instance, cost per add to cart, like a really fundamental metric that you would think it would have, it couldn't pull. And then it also can't look at any attribution. So if you're wanting like incremental attribution or seven day, one day, you know, any of that, it can't pull any attribution data. And then someone I was able to find said that it's only available to around 10% of accounts. I don't know where that number came from. Sounds right.

12:12Probably, yeah. You know, it's just a small percentage. Yeah. But I haven't heard anyone that does have access to it. I was actually only able to find one person on X that was actively using it. And they're the one that reported all these issues that they found. Now, obviously, it is new. So that's why they're rolling it out slowly because they knew probably add to cart wasn't working and attribution windows weren't working and all of this, which is probably why most people don't have access to it. But it's still a little disappointing that they announced it, this huge MCP thing.

12:43And then it doesn't even work. Yeah, I mean, as soon as it came out, obviously, I sent it over to the Breezeway team. I want to get their thoughts because one of the guys has like a master's degree in it in AI, whatever, intelligence stuff. The Breezeway team is incredibly smart. They're data scientists and smart people. So Breezeway, if you don't know, by the way, is our kind of meta ads intelligence platform that helps you make confident decisions in meta based on all your metrics. Go to Breezeway.co, free trial. So I sent it over to the team in the Slack channel, and they gave us their analysis of it, which is funny because they gave us what their thoughts were.

13:19And then you just kind of confirmed it's not really that great. Now, the thing I do go to is like we always have to consider where is it going? Yes. Because like obviously it's like the first time you launch the restaurant we love to go to next door, the first time they opened, it was like atrocious for six months. They had no idea anything. And I almost feel like this could have been like a knee jerk reaction to the whole meta situation. Meta is like full send. We don't care. Yeah. Send it out there, announce it, create a lot of hype, whatever. And so I think that where it is now is obviously now where it's going to stay. So it's going to get great. It's going to be able to do things.

13:49And I, it's hard for me to believe that meta is not going to only ever inject meta propaganda of what you need to do. That is the thing. Yes, yes, it is. And it's like as much. That was the issue with Manus. Oh, yeah, yeah. Yeah, it is the issue with Manus. Yep. Is like people, everybody wants the push button that there's no thought and we just trust the machine to tell us exactly the good stuff. Even Mark Zuckerberg wants that. Mark Zuckerberg wants that. And I want, that would be so easy, right? Yeah. And, but the thing is, it's just, it's not that easy because that's just not how, that's just not how the world works.

14:22Yeah. Because if it's that easy, everybody would do. So what happens? Simple supply and demand. The market always works itself out. Adam, Adam Thomas said that. Not the Adam Thomas that we know. Oh. Adam Thomas. I forget the name of the book. It's an economic book. But the market will always, always work itself out. Yes. And so like if it was that easy and you didn't have to build creative and it would just run itself and it would produce insane results and change your whole life, everybody would just start a business and be like, all right, meta, just here's my URL. It just sets the new floor. It doesn't work like that.

14:53It just doesn't work. So when people say that that's going to happen, I think that it's going to look different, but it's not going to be as, as like rainbows and unicorns as you might, as the Zuck might say. Yes. And so I think that how I really see this being valuable, because I'm like, how would I use this? Just time savings of logging into accounts. That's how I would use it. Like literally, I would just use Whisperflow. I'd go to Cloud and be like, hey, give me a rundown on metrics in the last seven days. So I don't even have to leave my phone. I don't have to log into a tool. So I kind of see it more as like we could probably build a Cloud skill or we could build something that would literally give us some visuals on it however we want.

15:26So I like the ability to like manipulate data to make decisions and see what decisions it might make. But we know that meta is going to spew its propaganda of what it wants you to do the same way it's meta pros. Yeah. Well, actually, I don't believe I don't believe that the MCP will allow that because you can't you don't like train an MCP unless it goes and searches the knowledge base. If it goes and searches the knowledge base, then it will spew propaganda. Well, if you're telling if it's if it's having you like give signal diagnostics with help in troubleshooting, like getting specific issues, you know, someone's gonna be like, why are my campaigns going wrong?

16:03And that is going to say, like, well, it's probably because you're not running a traffic campaign for us to do it. You know, it's going to say it's probably because you're not spending 10 times more dollars.

Limitations and Future of MCP

16:10Well, here's here's what I'll say. I went through and I kind of looked at what I think this is probably going to mean for people. And I came up with three, three key things that I'm going to kind of expand on here. Number one, what Josh just said, it's not going to magically turn your account into a six row as machine with the push of a button. You know, I install this MCP and now my ads are just working perfectly. It's not going to do that. It's also not going to replace a skilled media buyer. I'll talk about this a little bit in the future here in a little bit. And then it here's all it's going to do.

16:41It's going to level the playing field. Once again, as Josh just talked about, if everyone has access to it, then it's not special for anyone. So if you have access to it now, if you're not using it or you don't find a way to use it, then you are at a disadvantage. But if everyone's using it, everyone brings their floor up to the same point. So it's not like anyone is special because they have access to this. So here's here's the big thing that kind of all ties together. And this is how AI is working right now. Anyways, it's only ever going to be as good as your prompts, because if you just say, hey, go look at my account and tell me what's wrong.

17:15Right. It will only know what it knows. So it'll go in and it'll pull the data and it will give you something if it has to go and search the knowledge base because it doesn't know. It will only give you what is already readily available to everyone. It's going to require a smart media buyer, someone that actually knows how meta works to tell it like what is good and bad performance for me. You have to know what to actually ask it, what to pull from this so it's not going and pulling tons of data and then getting lost in the context.

17:50And then you also have to know what to do with the answer because, you know, it will spit out something that could be the right answer. It could be the wrong answer. And you have to know what to do with that answer. And that's still going to require a human brain. It's only going to make analysis easier. It's not going to just turn you into a state of the world class media buyer. It's only going to make world class media buyers better and faster. So that's the first thing. Here's the other thing that I think is going to be kind of interesting.

18:22It can't analyze your creative. It's not going to go and download all of your creatives and look at each of your images and your videos. All it has access to is your data. You know, you export a CSV. That's all it's doing. So here's something really interesting. You're going to say, hey, look at my ad sets and my ads and tell me what's working with it. And it's going to say, hey, your ad final V3 is doing really well. You should make more of those. But then it's not going to know the difference between ad final V3 and ad final V2 and why V3 is doing better than V2.

18:55You're going to have to know that. You're going to have to say, oh, in V3, I had this hook or this hook or I did this thing or I showed the product here. So this is going like it is not going to pull your creatives and tell you what to do with it. But I did think about this. I think this is going to become very important. Your naming conventions are going to become very important when you start using this MCP, because if in your ad name you have the hook type, the format, the angle, the offer, the talent, who's in the ad, what the copy is, what the structure of the video, if it's a video or the static or any of that,

19:30then the AI will be able to say, oh, any ad with this type of hook gets a ROAS above a 2.5. So it can then recommend to you, oh, you need more ads with this type of hook or with this type of talent. But you're going to have to give it that context in the ad name so it can actually say, oh, this is what's actually happening. Or you have to set up some sort of way for it to read the creative, which is extremely technical. Can I give my feedback? Yeah, what do you think? Here's my take. It can't analyze creative now, but it will be able to. And I actually, I think fixing naming conventions now makes sense.

20:04I don't think in 27 it will. Potentially. And here's why. I'll give you my take. My take is this. Google Gemini, by the way, everybody, I don't think that being a single LL family is really smart. I think you need to use them for the different things. So I actually do use chat. I use Claude and I use Gemini. Now, Gemini, dude, Gemini is really, really good. And we have a podcast, by the way, dropping next week, everybody, on a strategy on how to, we call it breadcrumbing on basically how to identify creative and then build creative

20:37using Gemini. And so Gemini is really good at analyzing creative. And so Gemini right now, it can analyze, statics are easy. Most stuff can analyze statics. But I would say that video is really hard because most will just pull a transcript and that's all they have. But Gemini will analyze what's happening in the video, whether it's short form, long form, whatever. And so I could see a world in the next, maybe in the next 12 months, maybe this year, but it might be next year where it's less about the naming convention.

21:07And when you're trying to analyze creative, the capabilities of the LLMs will only get better. And they will be able to pull in. It'd be hard for me to believe that Meta, who can analyze creative now and deliver that to the right audience going, running ads broad. It's hard for me to believe that they won't be able to analyze the creative for you and be able to push that context to something, right? So if they could push the context, I don't care what the thing is named. If I need to go build a new batch of creatives, all I care is what messaging is working, what ideal customer profiles, what formats, et cetera, are really working.

21:39And as long as I get brief of what I need, or I know what's within the ads, I can then use AI to create the brief based on the things that are performing. So I think naming conventions, I don't want to, I think naming conventions matter right now. I think though, in the next, next year, they won't. I think next year it'll just push the context of what's the best thing. And then you will just say, great, based on this, build me the briefs that I need to go build a new creative. Yeah. Then again, though, you are still trusting Meta to determine that for you. So not if you, not if you say, not if you tell Meta, tell me what's in the creative,

22:12tell me what, what you believe the ideal customer, tell me what is not what I should do. And then you could take what is, and you can manipulate, you could use Claude or whatever you want to then create what the next thing is. So you're, but you're saying Meta will be the one to tell you what is. Oh, in your, in your. If it's integrated, anything that integrates with Meta, you'd be able to just pull in what's in the creative. Or if it, or eventually Claude will be able to, where it's like, it could pull some kind of a file or something like that in and it's able, from Meta and it's able to do that. I don't think MCPs let you pull files.

22:43So that might be right now. That's what I'm saying. Like where's the, it doesn't allow, a lot don't let you upload stuff either, which is really annoying. I hate that. Which then I don't understand how ads, how you're going to be able to create ads with this MCP if you can't upload creative. Yeah. See, so that's a hole in the, that's a hole in this too. Yeah. Because Claude literally cannot upload files to Google drive to notion and it kills me because we use those integrations all the time. Yeah. So I, I don't think tangent. Yeah. And so I still think naming conventions are going to be really important because then the other thing you might need to do is run your ad through Gemini first to get that output

23:15of what's in it. And then you take that output, which is already possible today, but it creates another layer. So if you're not necessarily a technical person, you could at least put some, some naming conventions in there or keep your creative. That's another thing. Keep your creative readily available so you can upload it to these AIs because. You may never be able to pull it back from meta. You may have to pull it from wherever your file is stored, put it to the AI, let the AI take a look at it, tag it to what creative is actually running in meta and then run an

23:46analysis on it. So that's a, that's one thing. Here's another thing. This MCP, you know, the all knowing MCP will only know what's going on on meta. So here's a, here's an interesting thought. You know, you change your, your free shipping threshold on your website and it tanked. The MCP will only tell the AI that the performance of the campaign was declining. So then it's like, well, you know, it must, something must've gone wrong. You should turn it off.

24:18But that's not really the fix, right? Because it only knows what happened on meta. It sees conversion rate went down. What really happened was you did something on your website, but the meta MCP won't know that. So there's a potential here that people will start trusting the MCP, but the MCP doesn't know that your carts broke, that you change something on your website. Maybe you're out of stock of something or you change your free shipping threshold or, you know, whatever. So the AI is going to confidently tell you do this thing. You're going to trust it and do that thing when that was never the fix.

24:49So now you're stuck like in an infinite loop with this AI that doesn't actually know the full picture of everything that's going on in your business. Yeah. And this is already a problem. Like we're seeing, we do not recommend by the way that you export data from meta and then you try to have AI tell you what to do. It sees what it has the lens of performance of only that, right? It doesn't know these other things like Dylan's mentioning like issues on the website. It doesn't know free shipping thresholds and things are out of stock. If something broke in the car, like it doesn't know all of this data, it doesn't have this

25:21full picture. And that's why I believe it's going to be a little longer before we're able to truly get good paid media management automation with AI, because there are so many factors. It might tell you performance is like through the roof, scale, scale, scale, scale, but it doesn't know that you actually ran a promo sprint last weekend with a new product drop and you made $62,000 in the weekend. And that inflated all of the performance because of you through windows on meta. If that sounds like jumble to you, mumbo jumbo, it's because I just use a lot of big words. And so anyone who's listening and understands that knows what that's like.

25:53So meta is going to say, you need to scale. You need to scale. It's like, no, no, no, no. It's because we were in a promo sprint and it inflated the performance. And so it doesn't have all of that context. And that's why I think that do not right now, I just implore you, we have clients doing this and Robert said he's dealt with, he's had to like talk to a lot of people off the edge of like decisions they made based on what AI recommended from like exporting stuff and spitting it into Claude and asking Claude to tell them how to optimize their accounts. It's like, no, like there's, there's so many more variables to it. So I think it is still a while off before that happens, but I think it will streamline the

26:26way that you can manipulate the data that you need to make the decisions that are right. Yes. And then that goes back to what we were talking about. This is available to everyone. So it doesn't automatically make you a good media buyer because you have to understand what to do with the thing that it tells you to do, or you have to understand what Josh just said, that it has such a small window of data that it's pulling. So all this does is it raises the floor, right? We already talked about this. It raises the floor and the ceiling is determined by who knows how to use it best.

26:57Meaning a good media buyer will be the one that can use this the best. Now here's, here's the, the big takeaway. We've already been saying this for, you know, 12 months now. The creative is what matters. The product, the message, the creative is what matters because this is only automating the media buying process. Is my row as good? Is my conversion rate good? Is my cost per click good? All of this, but the people that will win are the people that are winning now that understand you need a good product.

27:28You need a good message and you need good creative. This whole thing is not death to media buyers. It's not overnight going to make like 10,000 good media buyers. It's just not, it's only going to make already good media buyers better and faster. So don't think that this is like the end all be all. It's going to make my ad account perfect. It's actually not. All it's going to do is make it easier to analyze, but you still need to know what to do with that, with that data. So hopefully this episode was interesting.

28:00If, if you enjoyed it, please share it with another e-commerce owner. It would mean the world to us. If you're on YouTube subscribe, so you don't miss, we are going to do a follow-up at some point when this MCP does come available, how we're using it, what we're doing with it. So if you want to see that when it comes out, subscribe. And if you're listening on Spotify or Apple, we'd love it if you'd leave us a review. We are trying to become the number one e-commerce business podcast. So a review would help us in doing that. Thanks for listening. We'll see you in the next one.