How to Build AI Workflows for App Marketing with MCPs & Skills

Of course, if you have a sound specific things um in mind that only you know only you do that that way um that's fine. Just explain it to AI. It's no matter like you would talk to ChatGPT, you would you would actually do the same here as well.
Welcome to another episode of ROI 101. I'm Roman from Tangent and today I'm joined again by Eldar. Hey Eldar.
Hey, nice to see you again. I'm Eldar. I'm founder of Hatchly uh based in London and I ship iOS and and different mobile apps and I do that with the power of AI. And today I think Roman will uh tell you more but we're going to be talking about skills and MCPs and I'm going to share you I'll share with you how I actually set up the whole process and how it all works.

Exactly. It's our second video with Eldar. Last one went great both in terms of uh your reaction, number of impressions. What we were talking about Eldar? Just a quick tease for people who haven't watched the video so they could watch it next time.
Yeah, so last time we were focused mainly on um social media and using AI to create content, specifically organic content and today video is actually kind of a prerequisite to that video in a way that you you going to use those skills and MCPs to actually replicate the results uh what I've shown in the last video. So today I'm just going to show you what's possible and then whatever you the viewer likes to actually implement in their stack, they can just take it um dive a little bit deeper and use it in their workflow.
Super. Uh you prepared the slides so just share your screen. While you're sharing, I'll

mention another important detail about Eldar. Eldar is Gen Z, and I find it like the way he solves a lot of problems. It's very unconventional, from at least for me, like coming from a different generation. So, yeah, just stay tuned and you'll see what I mean.
Okay, cool. Um Right, so let me share the screen. Um Yeah, awesome. So, today we're going to talk about agents, we're going to talk about MCPs, skills. And we're going to focus specifically on how they work um for mobile founders, mobile marketers, uh both for apps and games. Um we're going to focus on marketing stack today, but um we're going to cover it as a general concept, so that if you haven't ever used MCPs or skills, and you have no idea what that is, you'll understand how to actually implement that, and how to go and build something cool.

So, a bit about myself, uh I'm Eldar, I'm founder at Hatchly. Based in London, and I build iOS and AI products. Um you can check more information at eldaragayev.com. Right now, I have about 1.5 million users across my apps. Um I've generated across social media about 19 million views. Um eight apps I have right now live on the App Store. And um overall, I shipped more than 25 products. And I shipped them with AI and without AI, pre-AI as well. And the difference is massive. I'm going to talk about it a little bit later, but um yeah, we're we're going to I'm going to show you why the difference is huge, especially when we're talking about marketing. Of course, the development itself is now very simplified with AI, because you can just prompt your way to a finished product. Um so yeah, and I might actually tease you a little mobile game that I'm building

which yeah, I haven't planned it. I just thought about this right now. So
Is it a game or not a game?
Yeah, it's actual game. Um so yeah, and I've built it with AI. I spent some time and it's actually I don't know when this video will be out. Hopefully by that time the game will be out because it's it's very early very alpha. But yeah, you you're going to be the first ones to see it for sure. So yeah, and to continue so I also um a lot of hackathons. So I went to multiple DeepMind hackathon and you know, everyone is using MC business skills there. So the top players in the industry are using those and I think you're going to benefit a lot from that. If you want to talk more about this, you can check out my community which is um hopefully there's going to be a link to it. So and I'm also building a Meadows which is an operating system for agents to actually do the marketing. So let's now focus on

the plan for today. So I'm going to talk about the old workflow, what are the MCPs, how everything work. We're going to go over tangent revenue cat, Apple Ads, Meta, TikTok, Google Ads. And we're going to then finish it up with skills and how you can use Meadows on and other guardrails to actually drive through all of it. So just imagine you're back in 2024. Um you need to open multiple tabs, you need to go to multiple platforms to check different information to build some sort of spreadsheets to actually understand how everything works. Although tangent makes it you know, simple in a way, but you still need to go to an open tension. So, it's kind of you you still need to click through buttons, understand that you you've done everything right, that you've selected the the right filters, etc. And and I'm not even talking about like meta

ads, which are for most people confusing, right? And sometimes and same about TikTok ads, Google ads, everything. You need to be an expert in all of those platforms. Well, now it's Well, now it's much different. So, you can just use AI and that's it. You can just use your agent that you'll give special special tools, right? And we're going to talk about those special tools. So, what are the MCPs? MCPs is like a USB port for AI applications. So, imagine you connect a computer to your phone, right? Now computer can use your phone. That's the same thing with MCPs. The cable itself is the MCP for for the AI. So, how does it work? It's a protocol that allows any client like Claude, ChatGPT, Cursor, or Claude Code, or Codex to use other tools such as Tension, for example, RevenueCat, or Meta OS.

It help you to authenticate so that you can use your account, your actual account on Tension through your AI agent. And it's not it's not just about finding the information. It will actually do stuff for you. Meaning you can launch full campaigns without leaving your client. I personally like to use it in my terminal, but I know many people prefer to use it on online or use it elsewhere, and that's totally fine. That's actually more normal than my own approach, but yeah, the the thing here is that you can actually use that uh, with your clients. By client, I mean a Claude or ChatGPT. And and then you can do everything in one window, in one chat. You can ask, "Hey, can you actually check what are the my metrics on Tangent? Then can you go to RevenueCat and see my cancellations? Then can you go to Maddox and find what my my competitors are

running for organic TikTok?" And then you can "Can you create a new campaign in Google Ads?" All of that will be possible if you connect the right tools to your AI. So, yeah, MCP is hands and a skill is a playbook. So, they're two different systems. One is allow allows you to access another system, for example, Tangent. The other one it teaches your AI a lesson. So, these lessons can be saved as skills. Skills is just a txt It's not a txt file, but it's a md file, which is actually just text. It has some formatting in it. It has some different things, but it's it's just text essentially. It's the same way as if when you board a plane, there's a thing like that shows you the safety things about how to exit the plane in case of an emergency. So, that's the same thing. Imagine you have this playbook and you give it to your AI when you want it to do something. So, yeah, this is how it works. Typically

MCPs come already with skills. Nowadays, to be perfectly honest, I don't find myself installing a lot of skills because people who create those MCPs, they already already write scripts for you I mean, skills for you so that you can use them with those MCPs. And that's all actually you need to know. And then if you want some skill to be installed, you can find it online. You can search like skills for Claude code to do app marketing. And you'll find a lot of playbooks that different people use to actually do that or you create your own. You can just explain your own skill to AI, and if you don't want to re-explain it every 5 minutes, or every time you launch a new session, you can just ask it to save it as a skill. Another beautiful things about AI is that you don't even need to know how to connect an MCP or install a skill. You just tell it to do that, and it will use your browser to log in to Tangent to fetch an API key, and to install it on your

computer. So, that way you don't even need to actually, you know, ever log in to any platform and do it yourself. Of course, there is some guardrails, so AI won't actually log in for you. It won't actually, you know, maybe copy the API key, but it'll tell you in very simple ways on how to do that. So, the first one is, of course, Tangent MCP. It's a measurement layer. So, it helps you with your attribution. I actually have one one of your blog posts here. So, anyone who's wondering how to use that, they can check it out in your blog, new Tangent MCP server. So, you can create campaigns, you can delete them, you can update them. You can set up apps, server-to-server callbacks. You can ensure that SDK is healthy. So, one of the examples could be let let's say you have 10 different mobile apps on on two platforms. So, that's already 20 apps, right? If that's the case, you don't want to be in a position where

your SDK becomes unhealthy or something goes wrong. So, you can set up checks. For example, every week, your client will do a check on Tangent if all SDKs are healthy, and if something is unhealthy, they will let you know. So, that could be one of the examples how to do it here. Another examples is like a PLTV. I think it's predictive LTV, if I'm correct. Yeah, so predictive LTV can be used for, you know, planning your campaigns. So, in that way you can actually use it when you are planning a new campaign, you can actually understand, okay, so people who are coming from Meta are going to spend $10 on average per install. Fine, then we can we can spend up to $5 to to get to our target ROAS of or or P ROAS of uh uh two. So, that way we can actually um do that on Meta, but on TikTok it's it's the different PLTV, so we we do it differently. And you instruct your AI to

understand all of that and actually uh come back with different solutions for your marketing campaigns. Um you can do that without actually touching tangent data if you're worried about it. Just take tell it to do dry runs um before anything serious, right? So that you can ensure that whatever you're doing is correct. Um yeah, any questions so far? All all good?
Uh yeah, just a quick uh add-on here on the use case for uh skills. Uh what I'm seeing particular with the clients and like again, like new generation of clients, app developers, uh often solo or like some small teams, they will use uh skills to integrate the SDK. Like they they don't like they're not developers. They just go and I'm not even sure if they go to a terminal. They got just go to the cloud uh window and it's like

that's the whole thing. Um So, that that's that's the usually the first step. Another thing you can that I'm seeing people are using is that they don't go into the dashboard, they just do the audit. Like they have a skill that like yeah, told them this campaign work, invest more. This campaign doesn't work, pause.
Yeah.
And I I and I think it's the future. And I'm like I had um concerns, and I I think we also chatted with you like, "Oh, I'm giving the data to Claude or whatever ChatGPT." From what I'm seeing, even like banks are now like using this this way. So, you're either like behind uh and with your data that that you have never shared to AI, or you are developing many folds quicker versus the traditional approach.
Well, yeah, I agree 100%. I think that's the that's the case here. You either try

to be the first one, or you try to be the safest one, and then you might actually lose, and that's it. So, it's um sometimes it's not really a choice. It's just understanding that that's the trade-off you're going to make when you're choosing to use AI platforms. The good thing is in the future we might have a really powerful local models, so that way you can actually keep your data in. But for now, I would say there is no real alternative to something like Fable 5.1, uh which is the newest model at this time of the recording, uh or Astra model, the new one by uh Chat GPT or OpenAI. So, yeah, um I 100% agree.
Will you talk I haven't seen the full disclosure. I haven't seen the slides, but I'm curious if you talk later about like combining the different MMPs to achieve certain

results. So, I can use Tangent and Meta, so create Meta campaigns similar to the ones that performs well based on PLTV or whatever.
Yeah, well, I'm not going to I don't have like slides specific to like how to combine, because again, typically, you just say what you want. So, you just combine everything you have. That's the thing. So, whatever you use, just test if they have an MCP server. If they do, just tell your AI to install it.
Yeah.
Um and that pretty much is what it does.
By combining, I mean just connecting it to the
Yeah. Yeah. So, you you you just need to connect. I have a slide here on like how I use the stack
Okay. Okay. Okay.
altogether. But, yeah, it's pretty much um open AI and just tell, "Hey, can you install this MCP?"
Yeah.
That's the that's the
Exactly. I think that's the future. Like, it's not about using only Tangent

MCP or only Meta or only
Yeah, yeah, yeah, yeah. It's all together. Exactly.
Exactly. So, yeah, that that just I feel like it's an important point that I want to get across. But, yeah, yeah, carry on, please.
100%. All together, that's the way to go. In one shot, I can have 20 MCPs connected in my Posthog for analytics, uh my error tracking software for tracking errors to see maybe onboarding is broken because of that specific error. Or then I can uh see where the install came from through Tangent and understand, "Okay, may- maybe it's only people who have problems with onboarding are people who are coming from Meta, for example." So, those kind of insights you can only get if you have multiple MCPs connected uh to one AI client. Um and yeah, we actually talked about this on your blog post. You said that um you know, you don't need to leave the conversation when you uh when you connect multiple tools. And yeah, I well, it's hard to disagree with

that, for sure. Um the the entire stack is now moving to chat. I think also to voice because you can actually talk to it. So, what I like to do is that I wake up in the morning, I have my AI connected to all of the MCPs and I just ask it to do a report. And then I can ask it, "Okay, can you put it in HTML format so that I can open it in browser? I don't actually need to read everything in a chat." Uh so, that way is even more visual and I have all the information from all the platforms in one place. You can actually save it as a skill, like call it morning skill, and then just run it in one click. So that every morning you just run in one click. And again, to save the skill, you just tell AI to save the skill. It actually now knows how to save the skill, so you don't even need to think about how to do that. You just say, "Hey, actually I do that every morning. Can you save this process into a skill so I can run the skill and every

morning you understand what to do."
Yeah, that was my first reaction, like why don't you just schedule it? And then maybe you just like to talk to your agent.
Well, as Yeah, you can schedule it as well, yeah. Exactly. Yeah, even easier. So, yeah, the only important thing is that you need to schedule it and ensure you can actually follow up on that report if you have any any sort of questions about that. But yeah, you can definitely do that. So, the second one is RevenueCat MCP, which is the one here in the browser. So, it's essentially an MCP that allows you to connect it to RevenueCat and you can check all of the all of the things that RevenueCat is has online on actual platform, which is you can manage projects, you can manage apps, you can create products, you can connect it to different stores, you can create offering and packages. That's so much easier, especially

paywall management. Um that's also really good because it goes through their own AI-assisted paywall creation, which I'm not a big fan of, to be honest. I would rather create it on my own in my own cloud, but still even this one is quite good. Uh but paywall management or like everything else in terms of management is so much easier to do now. Because I actually don't need to leave my um chat. So I do everything here. I was just creating another I mean I was copying my app that I have on iOS to Android. And I you know you need to set up uh Play Console you need to set up products, subscriptions, you need to connect everything to RevenueCat, you need to create some sort of backend so that you can verify the purchases, etc. etc. I've done it in one session or one chat and it was done with MCP from RevenueCat and actually from Tangle as well, I think

and that connected it to Tangle as well. So yeah, you can do that in literally a few clicks. You just tell your AI to install it for you and it'll do so. Um the other one is Apple Ads CLI. So this is a custom one. It It now has been acquired by Roark AI and it's essentially a CLI that allows you to use everything you have in App Store Connect. So that means uploading new builds, changing metadata, changing keywords. It actually allows you to set up App Store ads as well. So I've set up full campaigns with Tangle and App Store Connect CLI because I can see the distribution of downloads, I can see which campaigns bring downloads, how it all works, and and I can actually update them as well. So that really helps me with everything related to um you know to driving traffic through App Store.

I've launched a new campaign just a few days ago. So far it's getting trials, it's actually working. So, um, you know, it's always done without even logging in into the App Store Connect. And, yeah, they offer so many things here. Um, how to authenticate. It's actually, oh, by the way, this is a really big um, big difference. So, this is not an MCP. This is a CLI. So, it's a command line interface. Yeah. So, what that means is that it actually needs to run on your computer because it uses API of App Store Connect. It can't be run on the server. So, because you don't want your credentials to be going to some third-party. So, you run everything on your computer. But again, if you have Cloud Code installed or Cloud Code Work or Codex, you can just ask it to install App Store Connect CLI, and it will do so, 100% guaranteed. And it will actually install

the skills as well. So, yeah, it has the skills, it has the CLI, and your, um, your Cloud will know how to actually run everything. And then, as you can see, there's so many use cases here. Um, you can, uh, localize the metadata, you can see the screenshots and media, you can add them, you can delete them. Um, I think you can also reply to different, um, new reviews of your app, um, and etc. So, you can do the full app management from one skill, uh, from one SCLI and a few skills, um, every day to help you keep your app up-to-date with the best ASO possible, with the best ASA possible, and with, uh, review management in place as well. And that actually, each of these things will cost you hundreds of dollars per month if you were to use some alternative solutions. Um, but this one you can use fully for free. I mean, only you pay for

your AI subscription, and then everything else you can do with those tools here.
A quick question or clarification. So, if I understand correctly, MCP and CLI difference is the CLI self-hosted at your machine and MCP is like encode MCP servers somewhere else. Is that the difference? Is that the only difference I would like?
Um so, CLIs were made by for humans, right? And MCPs were made by for AI. They both have can be self-hosted or exposed over um The CLI can be exposed over MCP as well. Um so, CLI is just a um command line interface, which means it's you write commands to to, you know, run some things. So, previously I would actually go to terminal and write some sort of command like this: App Store Connect builds, upload, app ID, package,

package version, version, uh build number, etc. etc. So, I would write those commands to actually for example, upload a build to App Store Connect, right? Now, AI can write this command in your terminal and do it for you because it already knows what exactly pass here. So, you can do that easily. So, it's a choice. It was a choice of the person who made this to use it as a uh CLI. Which is fair enough. It works. So, it's actually very easy to work with. So, why not?
I see. But, from the looks of it, it looks more technical. It's like the terminal commands versus the MCP where it's like human language.
Yeah, exactly. So, it was made for developers, but the good thing is you would never probably see those commands uh anywhere in the in the actual AI. When you use AI, it's going to run the commands. You just tell it, "Hey, can

you fetch um any reviews that were left on my app?" It'll just run a few commands, get the reviews to you, show it to you in the chat, and then you can just um do whatever you want with them, or ask it to reply, and then it will write the replies, and it will run a few commands to reply to those reviews, and do that as well. Well, how I use it is that every time I use Expo as well for downloading the builds. So, every time a new build is uh added to App Store Connect, um I will actually use these commands to attach it to new submission and send it for review. I don't actually log into App Store Connect anymore. I just ask it to do that for me. It'll write the the new update text. Um, it'll check my ASO, and it'll do that, you know, um in in a minutes. Um, and without me doing anything, actually. I just tell it what to do.
Got you.

All right. Um, so the next one is Meta Ads. Um, same thing again, you can use Meta Ads uh MCP server to actually create Meta Ads campaigns. You can do reporting, catalogs, AB tests, activity logs, everything. So, um yeah, it's works the same as a TikTok as well. TikTok has the MCP again, same thing. You can run campaigns, you can create campaigns, you can do everything you want. And same for Google Ads as well. You can you have a Google Ads MCP as well, where you can do everything here. Actually, this one uh yeah, it seems like it's legit a Google Ads um um GitHub. So, it's an actual official one. Uh they're updating it, so that's good. But yeah, you can do many things with that as well. If there is something missing for now, it'll it 100% will be added in the future, but for now you can use your local Cloud Code or Cloud Co-work

or CodeX to actually ask it to log in to your browser and use it in your browser. That could work as well. So, in case something is missing, just ask Claude to log into the browser and do that for you. Um I'll do that as well because it understands the web pages as well. So, it's very easy to see what's going on. So, yeah. Um this is the the stack, right? This is what we're talking about. You can connect all six of them in one chat so that all six of them working together. So, what that means you can actually read and write everything that you see here in Tangent, RevenueCat, Apple Ads, Meta, TikTok, Google Ads. Um yeah, and it the everything will work through natural language. You don't need to be technical, you don't need to say anything um you know, in a specific way. You can just explain your goal, what you want to achieve, and I'll do that for you. Of

course, if you have a some specific things um in mind that only, you know, only you do that that way, um that's fine. Just explain it to AI. It's no matter like you would talk to ChatGPT, you would you would actually do the same here as well. It might be weird at first because you've used to just, you know, logging in into the platform, changing it manually, looking at it, but then once you get used to that, you would never ever want to go back to the normal way. Yeah. So, now we're going to switch to skills and talk about skills a little bit. So, a skill is an instruction for an AI engine. It's like an um a lesson and um the AI you give it as a short description. For example, a morning skill, right? The one that we discussed that does the audit in the morning. You can give it a short description. Like,

for example, read this skill if you need to give a report and an audit in the morning for the user. It'll know then when you message the AI, it'll look through the skills and see if anything is relevant. If it is, then it will run that for you. It's cheap in a way that you don't need to use a lot of tokens for AI. You don't need to create folders, you don't need to create guides, doc files, etc. You can just write one MD file and that would be it. Uh one skill can be used across all of your different clients. So, you can use because it's just one file essentially, you can easily copy it to ChatGPT, to Claude, or locally use it in Codex or Claude Code as well. And it's the your way of doing things. So, I recommend actually instead of copying some other skills, try to use your knowledge and your way of doing things to create skills that fit your actual uh process. The reason why is

because if everyone will copy the same design skills or the same marketing skills, then the whole marketing and design going to be the same. And then you are not going to be in no way different from someone else running the same skill. So, that's actually a big disadvantage. It might sound and it might look like a shortcut in the at the start, but then once you realize the results are not there, you'll understand that all looks the same and you actually need to be creative. You actually need to be um you know, um creating your own skills. What do you think about this, Roman?
Yeah, before skills, I think like people were like sharing prompts like actively. Like, oh, here is a prompt to do XYZ. Like how would you describe uh difference now? Like of course no one is doing that, but they are doing skills, but if you can describe in a couple of sentences the difference.
Yeah. So there's no much difference to be honest. The big difference is that skills are saved for AI to be accessible at any time. So you don't need to copy

and paste the prompt. So imagine skills are prompts that are saved as files that AI knows short description of and can run the skill when needed. So it essentially will read that prompt and run that prompt when needed. So that's that's the whole difference. Prompt you will need to save it somewhere. You need to copy and paste it. So that's why skills actually were created, right? So that you can cut off this process, automate it all together. Because you don't want to copy and paste the same prompt, you know, every morning to run the same audit. You want it to be automated. So that's how that's how it's different.
Super.
Awesome. So some of the skills that I use is a grill me. I don't know. I actually don't didn't want to share it because it's such a good skill that I felt a competitive advantage when I was using it. So essentially it's a skill that tries to um grill you

on the things that you've discussed with AI regarding some sort of concept. So for example, you say let's test a new user acquisition angle. Um that's not enough for AI to actually go and do things because there's so many things it can do and you don't want it to be random. So if you run this skill, it actually come back and grill you. Which geography do you want? What creative hypothesis? What's the kill criteria? What does success look like? Where the data is coming from? What have you tried? So it's going to grill you through all the questions to get as much context as needed for it to actually start doing things. And it's very useful because I I want to make sure AI understands me to the maximum ability, right? So, these skills actually help you to to do that. And again, to install it, you just tell, "Hey, can you install Grill Me skill to Claude or to Codex?"

And it'll do so for you. So, that will work for sure.
Is it like uh business-specific or like it can also give advice or grill you on your life decisions?
Well, um I I haven't tried that one.
Well, I haven't I haven't tried it. I haven't tried it, but I I think it will. It will grill you through your life decisions as well. In a very business way.
[laughter]
So, it might work, but not the way you want it to work. So, you need to be careful with that, for sure. All right. Um So, yeah. Reporting, how it all looks. So, you know, you can create as I said um you can connect all the MCPs and then you can uh ask it to create an HTML file. The reason why HTML is so good is essentially this this page this page

this presentation here it's an HTML page, right? So, it it actually given me some cues on how to talk through this presentation. Um Yeah, so it's giving me some help here, but yeah, instead of using some other third-party dashboards, you can actually just ask it to create your own dashboard. And that way you can see the information in a way that you like because not you have that sort of information and you can create that sort of dashboards. So, that's helpful. An example could be here, you know, you can take the you have a TikTok, meta, Apple search ads, Google ads campaigns here. Um you can get those numbers from let's say tangent in one place, but let's say you also do uh some organic content and you have models for this. So, in a way to combine all of that together, you can just ask it to create the one HTML dashboard

combining information from all the MCPs that are connected. And you can save it as a skill to run it in the morning, during the day, and during once a week, um or whenever you like. Um so, yeah. That that that is a really useful thing to do. Um and um you can do that as often as you need because you can just create a skill and run it at any time.
Do you keep like recreating those HTML file or like Yeah, I got
[snorts]
I usually do because my each HTML file is a little bit different. So, I don't want it to be like super undeterministic in a way. So, for example, my question could be very different from time to time. One day could be like give me an audit on of my UA campaigns, another day could be give me an audit on my creatives across all platforms by creative distribution. So, it could it's different questions and different dashboards. So, that's why it's very useful to use like generative dashboards

because it's just being created for you for this specific day. Um yeah, so just to finish on skills, skills is just a markdown file. You have the name of the skill, you have a short description here, and you have the actual skill here. So, in this case, it's steps and rules. What to do and what to never do. That's it. So, that's how the skill works. It's the same way you give it to your new employee. And yeah, the last but not least, we have Maddox. This is the platform where actually this is the platform that was created as an MCP for agents. So, it's not created for humans. You shouldn't be using it as a human. Your agent should be using that. So, it's created that you never open the web interface ever again. You just connect it to your client, which is Claude, GPT, Gemini, or whatever you work with. Um and you have all the all the connectors um for your app that you need. You can

research your competitors. You can post organic content. You can uh connect to your App Store Connect. Uh I'd love to maybe in the future we can co-op with Tangent as well so that Tangent can be used with Maddox together. You can connect multiple different platforms through Maddox. And you can use You can not you, but your agent can use it in any way to grow your app, actually analyze it um from, you know, icons to screenshots to marketing campaigns to competitor creatives and everything in one place. It's currently in closed beta. So, if you'd like to if you'd like to access it, um send me an email or connect with me on LinkedIn. Uh but, yeah, this is the tool that is called a software for agents. It's a new new way of, you know, building software where you created for uh for agents, not for humans. Um yeah, and it has more than 400,000 apps. Right now, it's actually 2 million apps. With count 2 million apps, we know their screenshots, their icons. We store them. Your agent can have access to 2

million apps, understand their revenues, etc. etc. Um yeah, and it's free for humans. We're going to make it completely free. So, things you typically pay for, you'll get it for free. Um you can use it as a human. But, because humans are unfortunately inefficient, um it's not going to give you much advantage. So, if you really want to use it to the full of its abilities, you need to actually let your agent use it and it's a just paper use so you don't need to pay hundreds or thousands of dollars for it. Okay.
I have a question about my doors. Can you give me So let's say I'm a live coder. I created my subscription app that helps you with your nutrition, of course. How is the metal helpful for me? What
Yeah. So from [snorts] comparison perspective, it can compare

your icon to top 10 icons for example on the App Store for your niche. It can find how are your competitors running ads on TikTok, on Facebook. It can find your competitors organic content and it can find overall the best performing videos on TikTok or slideshows that you can copy and replicate. Um and then post it through metal as well. We're official partners of meta, TikTok, and Google so you can post your content to TikTok, Instagram, YouTube, and Facebook coming soon. Um and and you can do that all in one place in one chat. So in a way you you actually compare yourself to other to others. You see what your competitors are doing and then you use those learnings to actually create content. We're going to have skills for that and then those content that you create, you can send it to to different platforms to post it and then we again we're going to create skills so that you connect those everything you've seen here.

Um it will teach you how to connect your your AI to those MCP tools so that for example if organic content worked, you can actually launch it as an ad for on meta or something like that and then connect engine for all the analytics and attribution for that as well. So, it's kind of a connector of of all the things in one place. So, it helps you to get your app up and running and use agents as a operating, you know, tool or system for your app business.
Got you, man.
Cool. Yeah, so obviously be careful with API keys, but to be perfectly honest, this is AI made this slide. I would never put this slide because I have never had any problems with that. So, it should be fine. Don't worry about that. I mean just a little bit be careful, of course, but yeah.
Common sense.
Yeah, yeah, yeah. Common sense would help. So, how to action on this podcast after

this podcast? So, connect your MMP, so use Tangent. Then add RevenueCat, add other tools to actually everything that you use. Add it to your client, your Codex, to your cloud code. Add add platforms. It says read only. I do both, so go ahead and do both. Be a little bit risky, but it's fine. Common sense again, just check what AI has done and that would be enough. And write one skill. Just try it. Try it out how it works and see if you like it and then you can run it again and again. Yeah, that's it. Stop opening the dashboards. That's the last line of this slideshow presentation. It was created actually in about an hour with my eye together. So, it it's done a good job, I would say actually. It went well, so I really like it. But yeah, you can do that as well. You can actually have find presentation skill, use it in your AI to create those

kind of presentations. There hundreds of skills for that.
And as long as you know what you're talking about.
Yeah, absolutely. I mean you You to you need to actually be um um like you need to actually understand the the topic of course.
Cool. Like thanks a lot Eldar. Super helpful as always. Guys, give us a like. We will I'm happy to invite Eldar like for many many other episodes. So give us a like. Give us some ideas for the uh next episodes. Just leave them in the comments. Ask questions. Check out the links that we'll provide of course the Eldar products like Metal OS. His community we'll leave everything in the description. Uh yeah, I'm super excited where we This is like very start of this whole like MCP skills and I'm very excited what it's going to be like even in one year. I think it's going to be unrecognizable

uh market. Um yeah, thanks a lot. Then until next one. Thank you. Cheers.
Thank you. Bye-bye.