Connecting to Amazon’s APIs with MCP servers and Skills | Amazon X AI Working Sessions

Brett and I first met, I think, through Michael Maher at ShopTalk many, many years ago. And I've known Brett as the Amazon Cowboy for a long time. And then he kind of went hardcore into AI and that's why we're here talking. So without further ado, let me turn it over to Brett. Brett, please go ahead and get us started, please.

Yeah, thanks, Andrew. All right, let's jump in. My name is Brett Bohannon. I've been in the Amazon space for about 10 years. As Andrew said, you know, jumped into this AI thing really deep recently and kind of put a bet on MCPs and how people are going to talk to data going forward. So I'll be talking about MCPs, brand context, and the future of AI workflows. Um, this is one of my favorite things to talk about. I highly dislike slides. These slides were actually made by AI with my ideas, and they're not polished, they're not whatever, but it helps me stay on track for talking because I can go off track quite a bit. So we're gonna jump right in, um, and Basically, my goal is to show you what an MCP is, not to be afraid of an MCP, um, and also to kind of, you know, guide you in the right direction and then go through a workflow, um, like a live workflow walking through a whole entire thing using an MCP. Um, and that would be super fun. Um, okay, so in my tenure and working with brands on Amazon, a lot of it was you know, dashboards. One of my favorite things that I like to go back on is back in like 2018-ish or so, um, people were like, oh my gosh, I can go into this software and I don't have to log into Seller Central anymore. Uh, I'm able to kind of see my whole entire business. Um, and so I, I feel like we are at that stage again, and but instead we are able to talk to our entire business through different language models, uh, Claude, ChatGPT, OpenAI, Anthropic, their whole umbrella suite of products. So back then, um, you know, it's Seller Central, ads, dashboards, data tools, sheets, reports, things like that. So as an operator, you had to, you know, use these, pull these all together, and then create insights and then execute from those insights. Um, so like going forward and where we are now, you're able to reduce that, that distance between what I like to call it is insights to execution. Um, so basically you can get there faster. So the old way, um, you know, log in, click around, export. Compare and then explain. Um, you know, if you were Excel savvy, you have Excel formulas and all that good stuff and have super, uh, you know, tight, uh, you know, workflows that way. But nowadays it's— you can ask questions about your data. So I was thinking about this the other day when I was driving somewhere, and, um, and essentially it was, wow, like Back in the day, I would have to ask those questions and then kind of click around and figure out the answers to those. Now I can ask those questions and the answer will come to me. So I thought that was pretty powerful, um, and pretty exciting. And, and that's what I've been seeing lately is, uh, and talking to a lot of people is that that's where this next level is going. And I'm, and I'm going to talk about kind of the future of all this here in a little bit as well. But so back in the day, you're like, okay, where are my organic sales versus advertising sales? Like, that's a more complex question than how are my sales doing. So if you're asking a very broad question like how are my sales doing, you can log into a dashboard, um, whether that's Seller Central or software or whatnot, and you can find that answer. Now if you're like, what are my organic sales versus my advertising sales, you would have to— you could probably still log into a dashboard But if you wanted to dive deeper into TACOS, ACOS, and all that good stuff, you would have to stitch together some, some data. So that's kind of where MCPs come into play. You can ask these questions and then the MCP can stitch it together. Real quick, if no one's familiar with an MCP, it stands for Model Context Protocol. It's getting thrown out there quite a bit. And, you know, people are building MCPs on their data, which is great. Um, they should, uh, you know, and but essentially what an MCP is, without getting too technical, it's a bridge between the AI model and your tools and data. Um, it's pretty simple. What, like, why I've put a big bet on MCPs is back in, I think it was about February or March, um, there was, as these AI developments were happening with these different you know, Anthropic, OpenAI, and things like that. I was like, oh, there's got to be a better way to access this Amazon data. So, you know, I asked that question. I said, okay, how does this all work? And essentially it was like, okay, well, if these softwares have API access, I should be able to talk to that data within Claude or within ChatGPT or things like that. So that's how it stemmed. I, I was able to make— I call them adapters, but it was essentially wrapping these API datas and making local MCPs, um, for those particular softwares. Um, and it led to me having multiple different MCPs like within Claude. And then I had this aha moment. I was like, oh, this is going to be the future. Um, this is what people are going to start talking about. This is how people are going to start you know, conversing with their data and using their data. Now there's a lot of nuances that's involved with that, and, and it's not as easy as that. And they'll continue— like, we're in that beginning stage of all this, and it will continue to be— I don't know, there's a lot of little steps, but I'm not going to go there because, um, that's a topic for another day. So what are MCPs good for? Um, MCPs are good for ask across data sources. So as I was just talking about, different data sources, different asks. So you're able to ask it a question, it can pull back the answer. Um, it's good for not having to manually export your data. So if your data is connected, you could ask it a question, you don't have to pull that report. Um, it's good for monitoring changes over time, which is actually what I'm going to walk through. Um, this is actually one of the little sidebar here. This is one of my favorite things as a service provider because that's what I do too. I help brands on Amazon. Um, as a service provider, you have multiple things to look at, and one of them is making sure that all the listings retail— are retail ready and they're actually being shown to consumer. So monitoring changes is like one of the biggest things I personally believe is important for any Amazon business. So we're going to walk through a workflow on how to do that actually with an MCP. Um, so build repeatable workflows. So as you're looking through the workflows that you currently do, how can you reduce that time or reduce those steps? I talk to a lot of people, um, and let's just say maybe there's 6 steps, and this is what I talk to, um, often with people that are having their workflows. If there's 6 steps and you can plug in the MCP and you can use, you know, Claude or OpenAI or whatever, and you can reduce those 6 steps to 3, that's a win, and that's a very good initial step. Now everyone wants to go, I want to, you know, automate the whole entire workflow. Great, you can, and you will get there., but even if you can reduce it by 50, 75%, that's a huge win, uh, and MCPs help with that. And then, um, and then basically route outputs where work happens. Um, so you want to make sure that what you're asking it to do is actually valid and it actually does what it needs to do. Um, so that's the— that's one of the biggest things too. Okay, a couple different things when you're dealing with MCPs is how do you connect them. Um, this has gotten a lot better, um, in my opinion, uh, over the years. There's, uh, a couple different ways to do it. So it's a local OAuth, Marketplace, or custom. Uh, I'll just briefly go over this. A local is a little bit more technical. You just do have to do a block of code and you plug it in, um, to the Claude settings. So I'm just going to talk about Claude. OAuth is more, um, a connector base, so it's hosted web-based. And then if you have a login essentially for a software, um, it would, you know, pop up another screen, you would log in, and it's OAuth into that particular software and bring it all in. Marketplace is directly from like GitHub So essentially, if, if someone has an MCP, you can plug in that marketplace directly from GitHub and it will bring it into the connector. And then custom internal, you can make your own. So if you were into the more building side, you can definitely make your own MCP. So that's just a basic run-through of how to connect them. I'll tell you what though, you don't really have to worry about too much about connecting them. One of my biggest tips is if you're looking to connect an MCP, use like Cloud Code or Codex and say, hey, I'm trying to connect this MCP, can you help me out? And literally it will walk you through the whole entire process and it could probably potentially do 90% of it. Even if it says, oh, I can't do that, just push back and say, please try again.

Real quick, Brett, we had a good question come in. What is the difference between an MCP and an API? And I think that's such a great question.

Yeah, great question. Um, API data is, is used to pull, you know, that data into any particular kind of build-out that you're going to do. So you get the raw data from, um, any kind of software. Now, an MCP wraps that data and makes it available for the LLM model. So it makes it cleaner and makes it so it can interact with that data. So any kind of connector that you're going to do with the software to have it talk with Claude, you're going to have to use an MCP. Um, and then API, the data is a little bit more, uh, I want to say unclean, sort of say, for an LLM to talk to.

Yeah, the only thing I'd add there is that in an API is in MCP at the most basic level is just a wrapper for an API. And a lot of the more technical folks out there, especially if you're on LinkedIn a lot like I am, will point out like MCP is dead or that technology doesn't make sense because it's just a wrapper for an API. And at the most basic level, that is true. But good MCPs do a lot more than just wrap the API and make it a little easier to connect. We won't go too deep into that in this session. There's many more sessions we can talk to, to this. But as a specific example, when you're talking to an API, most APIs have good responses and bad responses, and a very basic MCP will pass through all of those responses to you, both good and bad, and leave it to you to deal with. And if you're not technical, that can be a nonstarter, right? Like you get some code response back from a system and you're like, okay, well, I don't know what that means. I don't know how to debug it. And you're sitting they're debugging it with Claude, right? A good MCP will already have follow-through and follow-ups for all of the most common failure modes of those APIs, and it'll take away all that pain for you. So the good MCPs will really, really just eliminate a lot of the operational burden of working directly with an API and maintaining that pipeline.

Yeah, that's a great point. And we'll go over briefly kind of some, some additions to MCPs. Um, you know, I'll talk about some brand context and kind of, um, good MCP kind of practices, sort of say, and additions to put on it. People call them skills, um, which is part of it, but, um, I like to call them harnesses, um, which have multiple different layers, which include skills and all that good stuff.

So yeah, adding jargon to the ecosystem, Brett. Jeez.

Yeah, yeah, sorry.

Artists, it sounds good.

There's, there's a lot, um, for sure, but, um, I like to try to keep it as simple as possible. Um, all right, um, so kind of going into the workflow I'm going to go over, um, so it's— this is pretty simple. So if you're building any workflow, um, within your business, you're going to want to connect the data, ask the question, define the rules, monitor changes, route alerts, and review the decision-making of that. Um, so I'll go over this here momentarily, but, uh, it's pretty simple, uh, in, in kind of the high level of it. Um, and as I walk through the workflow that I'm going to go through, we'll walk through each one of these steps as well. Uh, as you can tell, this was done by AI. So they have a little demo slot and all that. Like I said, I don't like slides, but I kind of have to stay on par sometimes because I can go off. Um, but I'm gonna walk through this right now. So what I'm gonna do is— this is, um, one of the adapters that I made for Keepa, actually. Um, if you're not familiar with Keepa, Keepa is a plugin. Um, it's been around for a long time, used to be free back in like 2016. They started charging and all that good stuff, but They got a lot of data, but it's one of the easiest ones I can do without showing you kind of, um, you know, content that I probably can't show everyone. Um, so took a sample list of some ASINs. I believe there's some acne patches just to kind of walk through, but, um, I'm going to walk through them here in just a second. And let me change my screen and we'll go through it all. All right, um, I have some pre-, you know, kind of prompts to put in here. So I'm just going to copy and paste those in. Um, again, I— the first setup would be brand context. I like to set this up so as you're in, um, Claude Cowork, you can set up a new project. Um, and then like I said before, this was started from scratch. So if you have an existing, um, any kind of, uh, folder with you know, context around that project, you can attach that. So I put in the brand context here, or the business context here, just so every chat can, um, can review that. I'm not going to do that again in this chat, but that would be like the first step. If you don't put in instructions, just be like, hey, here's my business, um, here's all the context around it. Um, and we did talk briefly about business context, so you can actually set this up, like, as a skill too. Um, skills get, uh, the term skills gets passed around quite a bit. You can have a whole skill sheet and you can have the, um, uh, you know, your chat agent update that skill sheet as things evolve. But for this purpose, we're just going to put it in as instructions and kind of rock from there. Um, okay, so here we're gonna have Or I'm going to put— copy and paste this real quick. Um, okay, let me copy and paste this text and get it in here. So I currently have the Keepa MCP setup Um, this was copy and paste from Markdown file, so that's why it's looking like this. So essentially what I'm doing is instructing it, use the Keepa MCP in read-only mode to monitor these lists, these ASINs. So let's just say these are ASINs that I have for my business. Um, return a concise Markdown table with one row per ASIN, include these fields available from Keepa. Um, so ASINs, product titles, short listing name, current price, coupon deal status, rating, review count, image count, um, and then has some reviews. So do not invent data, use the fields available from the MCP tool results, uh, right? Not available. I think Keepa has about 18 or so tools, um, that it has available. So every single MCP has different amount of tools, um, and that it can call for, uh, and that's what, uh, you know, those instructions entail. I already have the brand context, so I'm going to take this out of here and I'm going to say fake brand context— oops— is in the instructions. Okay, so we're going to plug this in and see what it does.

Come on, Claude.

See if it works. It's thinking.

Very nice. So you can see while we're waiting for this that Claude is actively loading the MCP that Brett told it to load, that Keepa MCP. He mentioned that it has multiple tools. It's loading all the tools from the Keepa MCP. You can see it saying finding tools. It's using that one called Keepa Get Product and Keepa Get Buy Box in order to start going through the, the things that Brett did. And this is a good time for us to answer, I think, this question that came through. Connect— can you connect Claude to the Amazon Vendor Central or Seller Central directly, or do you need other apps? And there's two answers here. The first is that yes, you can connect directly to Amazon Seller and Vendor Central through the MCP, Amazon's own MCP. It's something that they've just announced in very early beta at the Selling Partner Summit this month. But that said, they're very, very used to making these announcements and then those products taking a year to 3 years to actually launch. So just take that with a grain of salt. So Amazon is producing a lot of direct connection there. You can also potentially teach Claude or use Claude to build your own MCP that connects directly to the SP-API, uses your own authorization, et cetera, et cetera. And that's what a lot of companies have started trying to do is bringing that pipeline entirely in-house. House. But that brings us back to one of the initial questions that folks had, which is, you know, what, how is an MCP different from an API? And one of the key ways that it's different is that whenever there is an issue with one of the upstream data sources, say Amazon gives it, Amazon's APIs are down or they give you a weird response or they're getting throttled or what have you, that a lot of the well-built MCPs will handle all of that madness for you and kind of take care of a lot of the things that Brett was telling this Keepa tool to do directly, right? If there's blank data, it'll automatically tell you that it's blank data. It'll have a lot of really good tools built in to prevent hallucinations, etc., etc. So as you shop out there and I promise I'm not trying to sell you anything, as you shop out there, consider what additional layers the MCP that you're looking at adds to just dumping data from the API.

Yeah, I— Andrew Hamada. Yeah, I mean, yeah, I've had a lot of— I've learned a lot about Like, so I look at it like layers, like the data layer and then like the MCP layer, which is, you know, everything kind of in between, and then the operator layer. So, you know, kind of connecting all those dots. The data layer I've learned is really crazy. Um, like, and Andrew, you know that way more than I do, but you know, I had a meeting with a mutual person that we know, um, Nate over at Snowflake, and he's like a data guy like crazy. And he was like, you're doing more data things than you actually realize. And I'm like, oh really? Like, that's crazy. But what I, what I realize is, is data is data, but there's going to be so much to do if you just pull in raw data. Like, um, like what you do, Andrew, and what other data people do, it's like, it's, it's wild in a good way. But, and you know what, I think I, I also had a conversation with someone yesterday about about this is that, you know, if you want to build, great, you know, you're gonna have to— it's gonna require a lot. Um, but it's also leveraging people that have already built out that infrastructure, um, to use that data on top of is, is— could be the best option to do it, you know. Um, I'm just sharing what I've been learning over the past almost year about that.

So 100%.

Um, okay, so yeah, Keepa, Keepa was called. Here's the context right here. You can see which connectors it used here. Um, which is really neat. And then it came up with just like an initial pull request. So this is exactly what, you know, um, the monitor snapshot. So for instance, if you're monitoring any lace ASINs, um, you want to create an initial snapshot. So this is the baseline snapshot right here. Um, it's interesting, none of these guys are having a coupon or deal right now, uh, during Prime Day. That's, that's something that I probably want to check and see. Maybe it might have skipped, I don't know, or didn't pull that through. Um, but that's like—

I'm actually going to take a quick tangent here and actually just sanity check it. Why not? Yeah, it is Prime Day. That was my reminder to go check on my, uh, my wish list to see if anything dropped in price.

Oh, it didn't pick up Prime Day deal, so that might just be coming up as a price drop. Um, instead of a coupon or a promo on Keepa's side. Um, so maybe it doesn't showcase a deal, but it showcases like a price drop. Um, and that might, might have been just an instruction thing. So if you're like, okay, well, I want to track price drops instead, um, that might pick it up.

Um, but really I think the key thing here is, you know, bringing it back to that question that folks were asking about, like, what is the value add of the MCP in this context? Everybody, I think at this point is familiar with Keepa, has probably used the Keepa like UI or maybe their Chrome extension. Maybe you've paid for a subscription and maybe you've tried to use the Keepa API. And let me tell you, as somebody who has developed against the Keepa API, if you're technical, it's fine. If you are not technical, it is incredibly intimidating, hard to use, and potentially really easy to blow up your costs and blah, blah, blah, blah, blah. So, What the equivalent of being able to do this through the Keepa API through AI would mean probably a week of building and then basically forever, like spending some time every week, making sure that your Keepa MCP still works. If you use one built by somebody like Brett, then that maintenance becomes his responsibility. And if you use one ultimately produced by Keepa or somebody else, then it becomes their responsibility. So you can start to see where this adds a lot of additional help and just operational pain pain removed.

Yeah, and that's, um, you know, I think the GitHub community in general is maybe looked over for a majority of the people in our space. Um, I actually, you know, so if like you're using the Keepa adapter that I built, like if you see something that comes up, like you can put in an issue or a pull request on, on the GitHub and it becomes better for everyone. So these are open source tools, and that's the whole point, is that the community comes together and, and, um, contributes towards it to make it better. So that, that, that's a cool thing too. Um, one important thing right here is the buy box thing. So the buy box with the Keepa, um, has had problems. So this definitely is probably not zero. Um, so like one thing I would do here is drill into this and be like, the buy box is not zero, can you double check this? Um, for this intensive purposes, I'm not going to do that. I'm going to go over to the next stage real quick and just kind of talk about, okay, well, what next? You got this information, um, you got this deal. But that's why it's also important to look through this and basically be like, does this all look correct? Um, and then like I said, you can actually chat with it because the buy box field is available, but you might have to just kind of be like, hey, you need to check this buy box. I've had problems with the adapter. I thought I fixed it and patched it up, but it looks like maybe it needs another patch. Um, don't know. We'll look into that afterwards. Um, okay, so the next one is, um, I want to create a listing monitor, um, you know, situation. So essentially what I want to do is create a scheduled run for this. Um, so right here It just reformatted it, but we're going to turn this listing monitor specification, create two markdown tables, daily checks include price, coupons, deal status, buy box, whatever you might want to include in there, and then weekly checks. So right now, um, what I like to do, and this is— and it's also preset, um, you can have like, okay, what do I want my BSR fluctuations to be? What do I want them not to be? Um, so I created some workflows and skills around this within the tool that I'm building now, and it's basically like has some, you know, boilerplate templates where it's, where it's if anything goes above 20% change in BSR, I want to get flagged. Um, so you're able to use this and kind of say use it within your requirements and guardrails and whatnot. So essentially you're kind of prompting your own little harness, sort of say., within this. So, um, right here, uh, I don't think right here it says, you know, I want to be like, you know, create a schedule for this, for the weekly and the daily. So this way it's running on a schedule, um, and it's operating without you having to prompt it into any chat. Um, right now it's— we're running a little bit on time, but I'll go over quickly. Uh, it's just a markdown file. So I'll give you an example of what I do. I have reports that go into my Obsidian vault, and then I have a daily agent that runs through those reports and gives me a report every morning that I look over. You can do it differently. You can have Slack notifications, you can have email notifications, you can have any kind of notifications that you want within this realm that is connected with Claude. So you can go here and essentially be like, okay, I want to, uh, do Slack connector. Can you, um, you know, give me these reports on the daily basis only for things that are flagged, or for every report, and send it into my— this Slack channel? Um, so you can do whatever you would like to do in regards to notifications for it. So right now it's just going through these instructions, um, of the daily and weekly checks and creating the schedule for it.

So really, the, um, you know, while we're waiting for this, to me, like, the, the, the acceleration here has happened in a number of different ways. The first is, um, through the documentation of all the context and whatnot that you ever need instead of needing to do that on a per-chat basis or on like a per-project basis. You can now capture all of that forever in one place, point a tool at it, whether that's Claude Co-Work or Claude, uh, something else or ChatGPT or what have you, and all that's portable. The second is you also have a bunch of instructions, uh, both in markdown file specific to this, but also we can see that this tool is actually actively creating its own markdown files to teach itself going forward and capture what it's learned here. And really what the MCP has done in this context is helped you move past a lot of that initial operational setup because instead of needing to think about how do I connect to this API, how do I validate what I'm getting back from this API, how do I troubleshoot when it doesn't do what I want it to do, et cetera, you've moved on to the actual creation of this daily price check, which I think is what everybody on this call really cares about. Becoming an expert in rather than becoming an expert in actually creating workflows.

Yeah, yeah, 100%. Um, I think the, the goal of everyone is to save time, you know. Um, that, that's been my goal. Um, you know, I've got a family, I got 4 kids. I want to spend as much time as I possibly can, which is funny. Inversely, I've actually been more busy, but, but, uh, um You know, you want to be able to be like, okay, well, what do I do on a daily basis? Or have your employees, you know, 10x their output, you know? Like, um, that's, that's another big thing is like, how do I take what I'm currently doing and make it better and concentrate on the things that actually drive business, you know? Um, as an Amazon operator, it's like, you know, and, and someone that helps brands, it's like, I want to work on the strategy. I love that part. You're like, how do we get from where we are and keep growing and growing and growing, you know, you gotta, you gotta click a lot of buttons in that process. But if you can have something like this where it takes a load off of you so you can continue that strategy or you can continue that growth, to me that's a win. Um, so yeah, that, that would be my biggest takeaway. Um, so yeah, it just created a whole Markdown folder. Um, you know, has a scheduled daily run. Um, next thing you can do is you can put in this schedule. So you can essentially just be like, okay, I'll just do this real quick, and then Andrew, we can jump to questions, or you can cut me off whenever you want to.

No, you're good. Keep going.

So it should pop. And I'm a horrible speller. Um, you know, I saw, I saw the guy from, um, who created OpenClaw. He's like, he commented on someone and he's like, you shouldn't have to spell when you're talking to these, to these LLM models. They should be able to interpret all, all, uh, all spelling errors. So I'm like, oh, that's good to know.

Yeah. Although then it, then it depends, you know, like you start to ask yourself how much, uh, how much token burn am I willing to spend on that, right?

Good point, good point. That's a good point. Um, yeah, so right here, create a schedule. So if you're not familiar with creating a schedule in Cowork, um, it will create one. Uh, it will arbitrarily do a random time. My recommendation is you do a time that you're at your computer, unless you want to keep your computer on, unless you have always-on, you know, machine. But I'll just, for all intents and purposes, I'll just create a schedule here that's creating the daily and weekly. Um, right now it's like I mentioned before, it's just going to pull in a Markdown folder within the chat, but you can have those go anywhere you want. Um, I'm not going to go into detail about that, but you know, if you live in Slack, you know, just connect your Slack right here. Uh, and then say, hey, I would love these in Slack, and it will do it all for you in Slack.

That's awesome, man. Do you mind as a last closing thing going back to the screen where you were showing us, or maybe it's in your presentation about all the different MCPs that you're working with or that Helm you build connectors for with Helm?

Yeah, yeah.

See, because really here when I think about you know, the summary value of what something like what Brett has built here, you know, and bringing it back to all those questions about API versus MCP and how is this saving me time, etc. Brett has gone in and built MCPs for Keepa, for Datadive, for SmartScout, for auditing Amazon Catalog natively, and all of these things are turnkey usable by you. And instead of needing to go and think, how do I then go and develop API connectors to go and talk to all those tools? Where does that data then live? How do I stage it? How do I put it into my analytics tools, et cetera? Like everything that I just described was a like $150,000 a year salary minimum full-time 5 years ago. And now it is, trivially easy for basically anybody on this call to start doing that kind of integration and bring all of those data sources together directly into Claude, Claude Coworker, whatever other tool that you want, and to start going through that automation that Brett just walked us through.

Yep. Yeah, yeah. Helm is Helm. I built out Helm to kind of like have the harness be, be the main thing and the feedback loop around it, around the multiple MCP calls. Um, so kind of where this is all going, we're gonna, we're gonna be in that, uh, season where you're gonna have— you have multiple different softwares you log into different. Now you're going to have multiple MCPs and you're going to be talking to multiple different MCPs. Now the thing is, is if you don't, if you don't direct the— your chat to go to a certain one, it might take a question and it might call the wrong one. Um, and I've seen it firsthand, and that's what like made me develop Helm, um, is it helps me on an operational level work better. And within that, I've put in workflows that I've done over 10 years. I put in, you know, skills, brand context templates where essentially you just have to tell it a couple things and it already uploads all that into that. Um, basically just make it easier to operate with. Instead of, you know, redoing that whole entire workflow that I just did.

100%, man. And that's why I'm excited, why I was so excited to put this event on for this audience, because AI, for better or for worse, is— it feels like almost purpose-built for people like us. Yeah. Where you have years, probably decades of operator experience built up. And the bottleneck for you being able to deploy that creatively has always been I only know one technical skill or a handful of technical skills, if that, and that's not my day job. And now suddenly all of that barrier has become very, very easily surmountable. So, uh, thank you very much, Brett. This was really, really helpful for me personally, and I'm sure for a lot of folks in the audience. We've still got a few minutes here, so I'd love to turn it over for questions. Um, the first one that we got, um, probably related to the catalog monitoring example that you gave, or the listing optimization example, uh, came in anonymously. What advice do you have for people with really large catalogs of 1,000 or more products?

Uh, that's a big— that's a broad question. Um, if you're trying to monitor them, I would pick, um, the ones that matter the most to you for that particular, like, scenario. Like, if you have, like, you know, A-tier ASINs and you have, let's say, 20 or 30 of them, 50, 100 of them, um, you know, monitor them that way. Um, it depends also on what kind of Keepa subscription you have. Um, their thing is token-based, so you can spread them out throughout the day, um, and your tokens will refresh within it. Um, you can even send their API docs into a Claude code or a chat and be like, hey, can you explain to me like this? And can you— here's how many ASINs I want to monitor. Like, could I spread this out? Um, you know, what would that look like?

And it can give you a whole entire plan for Yeah, but basically the main point there I think is that it costs and complexity do scale directly linearly with async count. If you're going to work through something like KIPA, what I would say is if you're not already registered for it, our next session actually addresses catalog management at scale with AI. So hopefully you get more specific and direct insight there. It's specifically about, from our friends over at Autopilot, uh, who do that as their day job.

That'll be awesome. Yeah, Andrew, I actually have some free skills if the audience wants them, if I'm— if that's okay with you. Um, this, this pack, um, you can check it out right here. It's a 5 skills pack. I've gotten some absolutely fabulous feedback on it, um, and the whole purpose of it is just straight value. So if you want to start with like one of those workflows, this is a great starting point. And I don't, I don't give away stuff lightly. I'm, I don't, I don't do that stuff on LinkedIn, and, and I make sure the stuff that I, I want to provide is of value. And so I, I feel completely compelled to share this because it does provide value. So play around with it, um, you know, check it out for free, um, and if any questions, let me know.

That's huge. Thank you so much, Brett. Uh, one last question that just came in, um, in addition to acknowledging that you've just done a lot of cool stuff. Um, security— should folks have security concerns with using tools that have been created open source, whether by you or by other folks?

Um, yes. Um, so security comes about when you have quite a bit— you know, there's always that involved in it. Um, there— I can go really deep into security, but if you're worried, you can I'd literally be like, hey, is there any issue with this particular— like, go into Claude Chat or Claude Code, you know, Claude Code Chat, and be like, I want to check out this GitHub repo, um, is there any security issues that might come about with this? And it will go through the whole entire code and it will tell you, yeah, this looks a little funky, like, stay away from it. Um, I think as I think that comes about as there's a lot of contribu— you know, contribution that comes to those particular open source repos. The ones I have, there's not many, and I, I go through them to make sure that there's nothing bad that comes into them. Um, so yeah, I mean, my code is, is fine. I understand hesitation and security around it. You always want to be careful when you are working with an open source. You want to if you're not a technical person, you definitely put that through, you know, a code check and make sure everything looks good.

One of my favorite tools that we used to use internally at Amazon when we were writing docs is in the PR FAQ, one of the FAQs was always like, what is the worst headline you can imagine someone writing about this product for Amazon? And I've used that same mental model when I'm inspecting things with AI. So I'll ask AI like, hey, imagine that you are trying— you're a security researcher. Like, what are the absolute worst things that you can see happening because I use this tool? And it's going to scare the crap out of you, but probably in a good way. And as you do go through a few reps with that, you'll start to get a sense of, okay, here are problems that I really care about. Here are problems that I care less about. Here are problems that I already have solutions for.

I love that. That's awesome.

Brett, thank you again for sharing that free Skill Crate pack. Thanks very much, everyone.

Thanks, everyone.

Be well.