Ahrefs MCP: A Practical Walkthrough of 5 powerful use cases

Okay, live stream set up which means we can start the webinar. Hello. Hello. I see that lots of people are waiting in the waiting room. Welcome to today's webinar. I want to quickly set up a poll here uh while we wait for everyone to get in uh Uh, I just want to create a quick poll. How can I do that? One second. Okay, I'm just going to create a single question. Yeah, welcome everyone. You can say hi to each other in the chat uh while I quickly get this poll set up here. I want to run this poll. I like to get an idea about how familiar you guys are with AI NC and NCP because I have a section at the very beginning uh that talks about how to get it set up for uh for your uh own account. And if you are you've never used it before uh and and you're very unfamiliar or most people are not uh very familiar with the tool then it I will go slower in that portion basically. Okay. And there seems to be an equal split between people that have never tried people have tried a bit and people who are active using. Okay. Uh so I think I I'll at the beginning I'll share that I feel that if you've already been actively using AI or have connected a bunch of MCP servers together um today's content might be a lot of it might be going over the basics again. Uh, but I would also hopefully you'll still learn something new and if you have any particular questions about the HFSN NCP, I am here to help answer those questions for you. Okay. Uh, meanwhile, uh, thank you for everyone that helped to answer the poll. So, we have currently 114 people answered out of 774. Okay. Awesome. So there is basically a three-way split between people have never tried HFSN MCP and AI, people that have only tried a little bit and people who are actively using AI. And I think it's very important that especially even in my own journey uh that it's never too late to give uh AI a try uh to help automate your workflows and and like create like exciting things for you. Uh in fact I would say that uh it AI has really never been more capable, more powerful and less hallucinated than it has ever been. Um however I mean there's still like challenges of using it. So we can go over that uh as we go through today's presentation. Uh and yeah, some Tim uh Perslo said just because we're using it doesn't mean we're doing it right. Um, that's true. I think I would honestly put myself in the same bucket, but I but still I would encourage hopefully after this webinar, you'll be encouraged to give it a go and try even more of what you can do uh to get uh AI working for your workflows and uh like improving your day-to-day tasks. Um so I will go ahead and uh without further ado uh start today's webinar. So welcome everyone to today's webinar. Let me share my screen. Let me share one second. Put myself in the very teeny tiny corner here. Yeah, that's a quick Okay, so let's go all the way back and present. So, so uh yeah, I am using Gamma for for my slides because it helps me quickly update stuff uh that I want to add to slides. But yeah, so anyway, uh welcome everyone to today's HS webinar. I'm your uh host constants and today we are going to cover five useful use cases of uh HFSMCP. Uh if you have a question so I'll be sharing the main content of today's uh presentation. Uh depending on how fast or how slow we need to go I may have to skip some portions but don't worry at the end of this webinar and and the email following up to all of you guys that attended um I'll be sharing all the resources that we shared today. Uh and and I'll also of course stick around for a Q&A after today's session is over. Um if you have a question though because we want to make sure that we cover as much as we can today, please use the Q&A function in Zoom. It helps you um that you'll be able to find it under the more I think in your toolbar. Uh there's a Q&A function to ask any questions you may have along along the the webinar and then we'll answer them at the very end. Okay. Uh this webinar is also being live streamed to YouTube. So if there's any part of the uh today's presentation that you'd like to watch again uh that is your place to to watch it afterwards. And that replay will also be available uh at the end of today's uh present uh webinar. I'm going to go ahead and end the poll here. by the way. So, thank you for everyone who participated in the poll. Helps me figure out like how like where how I should go about with today's presentation. So, um so without further ado, let's begin. Um I want to give a quick disclaimer. I'm doing my best to share my learnings, but there's still a lot of stuff that I personally am like learning through um doing NCP. So if you have any for example technical questions or like things are not working for you or like uh you see that some API call is not working uh or like some NCP integration is not working uh I won't be able to answer here in this webinar a because it's very hard to sort of answer individual like troubleshooting for this and B there's probably there's a lot of technical questions that uh it would be better to um talk to our support team. Um today I want to focus generally on how we see um how you can use HF's MCP as well as certain best practices that I can share um from our time using it. Um I hope that what I'm about to share today will be helpful in in letting you guys learn how to use uh HFS MCP. Okay. So let's go forward. So let's start with the basics. What is MCP? So it is uh moto context protocol. that's what it stands for. Uh and it's a standardized way that AI can basically talk to a third party uh tool or program to either extract data or perform particular tasks. Uh in the past like um because AI is sort of [snorts] not deterministic. they will keep trying to figure out and and generate um ideas in order to access data on a website or pull or um like for example do things like uh access API and then like maybe the API URL like changes over time it's not correct uh and basically trying to connect with all these different apps took a really long time the structure is always different so in order to solve this problem this is what this uh protocol is about it's basically to um standardize the way that AI can or offer a standardized way to the AI to access and connect with different tools. Okay. Uh and HFS also has our own to which you can use to connect with our API. Um why okay so I've kind of covered basically the reasons why um MCP is is useful because like there there are also a lot of hallucinations. There are a lot of like when you want to connect APIs like traditionally there's also a lot of errors and if you are a programmer if you're not a programmer you don't have technical knowledge basically you would struggle figure out what went wrong uh and today now you can use AI to basically do all of that for you can just use plain text English to be able to uh connect with different apps without understanding the technical knowledge be behind how an API works uh and and that should be like a superpower especially for many of you guys uh who may not have technical knowledge uh around APIs and stuff. So um one second uh so how to set up the HFS MCP. So this depends on the AI that you're using. Uh I don't want to run another poll. So can people just sort of message in the chat which AI you guys are using today? Typically I'm seeing Claude Claude co-work chat Gvt Gemini Enterprise okay claude okay so today I will cover so I'm just going to quickly open our documentation Okay. Um we have some general setup guides a lot of it for like different claude programs but also uh chat GPD and lovable and ne um but I will show you overall how um it like connecting MCP connector usually works and it's it's quite simple actually uh but if you want more details as to how go uh how to go about it as well as the options to do it. So sometimes if you want to um do like uh connect your NCP server only via token not through authorization you can see the installation steps to do that. So while I'm here I will just share for but because there are quite a number of you that have not used MCP before uh please um based on your tool of choice check out the documentation here but I will quickly show you like kind of like uh how it works. Um so uh for just so that you guys have a good idea about how how quickly you can set that up for you. But just note like especially if you're using claude um sorry one second pasting it again. So if you're using claude or any um AI platform you need to check if you have a lot of these AI platforms like require you to have a paid subscription in order to use the AI connector. Um I did provide uh a free uh platform today just so that people who have not paid for any AI platform you can follow along. Um but I cannot guarantee I I'm not going to say like oh which platform is more useful or or or advocate for any of the platform so far because I think um that's this today's webinar is not the place to do it. Uh but either way you can see we offer documentation to help you connect with a lot of these different uh AI uh platforms for HFS NCP. And um to quickly show you something else as well uh it's important to note that your plan has NCP uh limits. So uh sorry API um limits. So in order to check that you can go to limits in usage in your HF's uh account and it's all the way down here API units per month. Okay. And so um it you can see that it is the units are shared across direct API access MCP server and other like uh HDS connect uh integrations that we support. So if you're using the API directly and using the NCP will share from the same limit. Okay. Uh and the only way to increase that limit is to upgrade your plan basically. But uh but you but every plan should have some and recently we also announced that the direct API is also accessible uh for every plan include um starting from light. Okay. So that's also like something you should keep in mind like you should be able to see this limit here basically and follow along to see how much you're spending. Okay. Uh all right. So adding adding let me change the screen that I share so that I can show you the this example that I'm sharing. I think first I will do claude. Let's do claude. Something is not working. So let's go. All right. It works. Now we're back. So you'll see that uh you will be prompted to log into HFS. And after you log into HFS, you have to pick your workspace. Uh in this case, I'm going to pick uh this one here. And then it says allow access to your workspace. And you click on allow. For those that are a bit more senior, bear with me just so that everyone can follow along. So now that you see now that I go back now I'm going to go back into cloud code and show now I'm going to go back into cloud code and show what I can see here is that you see all these connectors that are available. I just uh managed to connect Hrefs here, right? And then if I go back to one of the chats, important to check is that I want to make sure that the connector is turned on. Okay, so make sure that your connector is turned on before you uh request your uh the you want you want to request API um your AI because uh some if this is not turned on basically it will not work. uh it would not be able to access the MCP server. Okay, so that's basically how you connect and like I mean it was a bit complicated but that just now but you saw that when the workflow works it's just a few minutes to get that connected. Um I'm going to now switch over back to our presentation here. So this should be ready now. Let's go back. uh tracking units consumption I shared already. It's in your uh account settings over here. Uh there are a few things to check. One is like if you want to check your limits, it's under limits and usage. And this is across all your uh API uh tokens as well as across all your users. You'll be able to see in the summary of like how many API units you've consumed for the month. So this is the space to do it. It also has the option to like uh or in this play in this case because it's enterprise plan if you're on enterprise um you can purchase like additional credits if you need uh units if you need. Now if you want to look at which key is being all right let's not on this page I'm going to log into a different workspace here so I can show you 875 So if you in your account settings, you should also be able to see API keys. So now you can see all the different keys and what they're being using uh what that um whether they're used for API directly or via MCP as well as the units they've consumed and information about like adding and expiring. So this is your place here to generate new keys if you need to as well as uh like uh delete old keys and generally try to find which key uh is using the most credits and stuff like that. Okay. And yeah, like uh some you can see like if you're running out of your credits really fast, this is your I mean your your units really fast, your API units, you can use this page to narrow down which is the API um key that is using up the most units. Okay, so that is useful to know. Okay, let's go back. I'm going to close this for a minute. Uh so we're going to cover five ways to use NCP. The first is like your very basic use case where I just want to ask AP um AI to do some keyword research for me uh and just simply ask it uh to do stuff based on uh based um by by pulling HF's data and presenting me with a list of different options of uh keywords that I should um and and explain a should explain to me how they are useful or um worth my time creating content around. So here's the first one. So uh let me I'm going to uh show this with two um two AI platforms. The first is Claude and the second is Mistro. Okay. So let me switch over to Claude. Uh I I did this uh research over here where I asked it to uh find keywords related to skateboard and find the top 200 related terms and top 200 search suggestions that have at least a search volume of 200 and have a keyword difficulty 30 or lower that were first discovered six months ago or later. So I want to have so in one like like sentence I wanted to ask for uh keywords related to skateboard. It may not necessarily include the word skateboard. um that are some of them may like from search I think uh search suggestions means that I'm also interested in looking for uh keywords that pop up uh when people are typing a search query not necessarily in the HF's database and uh I want it to be recent um terms not uh terms that are necessarily old. So I want to look for trendy topics. So then I asked AI to group the keywords similar meanings together and and to provide an appropriate group category for each keyword and then I asked it to pull out a CSV file um for all these suggestions um providing all the information that I wanted to uh get for every one of the keywords and then summarize the findings. Okay, just as an experiment, if you'd like to follow along, I will go ahead and paste this prompt here in the chat uh here, which if you are have your AI open, you can then use this to to go um try to get the same results yourself. Um, so then if you would like to follow along to see what happens with uh what AI does is that you can actually open this up like they'll have all these steps here and it'll show you um what its logic is but also it would be able to share. Wait, not this one. B here here you go. It shows you what API it pulled as well as the results that came from the uh API uh results as well. This helps you to debug. So sometimes when AI goes like oh you know I was not able to find the information or sometimes AI is a bit sneaky and it would like give you like present you with like information but like the MCP connector didn't work for some reason. uh and your only way of like figuring that out sometimes is that you have to check on these different steps to see whether or not results came back or not. Okay, so this is helpful for debugging issues. So then you follow along with like what AI is trying to do. It basically got like a different uh suggestions for the seed keyword that I entered in uh in the original prompt and then it said I I came up with a bunch of different suggestions. It then built the uh CSV and then uh basically give a summary of of the results here and uh it provided me with CSV. So I can open this up on the side in claude. So I can see the results myself just within claude so that I have an idea of like what are the keyword uh suggestions that they have provided for me. I then like uh asked it to in a subsequent prompt I asked them to group all these keywords into word clouds and it did that. So basically it helped me understand uh how to look at different uh skateboards uh skateboard related keywords based on like kind of different categories that they semantically undecided. So some are related to products, some are related to components, some are related to a particular location and stuff. And uh then I also noticed that it's quite a number related to DG K and like you know this is your opportunity to then see these at a glance. It's like okay I can dig in to see what this uh brand is about. Yeah. So different uh like customer uh segments related to the skateboard. So like ba b b b b b b b b b b b b b b b b b b b b based on this keyword query, I've um shortened down my keyword research efforts from looking through all these keywords uh individually in a CSV like before and now I can speed that up uh by understanding what uh categories of content I can write for uh my seed keyword. Okay, so now it's really lessons and so uh history of skateboarding. So like this gives me a great idea of like uh overall what are the different content uh categories that I can target uh based on the keywords that I have searched that are of recent times like they're they're recently um that are keywords that were at least discovered six months ago. So instead of pulling this data through uh our platform and then like exporting it and then doing a filtering multiple times, I can now do it in a single prompt and uh ask AI to help do the analysis for me. Okay, so this is like your very basic use case and I promise that I'll be able to show you an example with a different platform and I will. Let's let me do that just very quickly. Uh can you guys see my screen? Okay. So um I tried to do it in Mistro. So why I I've suggested Mistro is because um unlike other platforms uh AI platforms, Mistro provides you a way of doing uh like connecting MCP servers without a subscription. I'm on the free plan. So if you are a skeptic of AI and MCP servers and you really don't want to pay for something upfront necessarily, you can try um using mist basically. Yeah. So, um, if you if you can if you if you'd like to give it a try, then then like I would this is like one of the AIS that I know that provide NCP connectors for free. And so, uh, you notice that different AIs will present findings differently. So, in this case, like you have to download the CSV file and then it gave its summaries in uh, different lists here, but you can see some of the keywords are similar. So, it's pulling from the same data uh as you would expect from the HF's MCV connector, but it gave some analysis about like which keywords were were higher competition or lower competition, stuff like that. But, and you can also see the analysis is much less. So, um different AIs have different pros and cons. Uh either way it is a like a lot time saved compared to if you were to do this manually especially if you are not uh even if you're like building a script or something like an AI saves you a lot of timing doing this analysis uh compared to before. Okay. So that's for uh this first use case. so let's go on to the next one. Uh finding backlink building opportunities. So I want to show this via claude code. Um claude claude provides a way for you to run claude code from the uh uh run AI via uh I think I will try to use terms that I understand. I may not get it right but um cloud code essentially allows you to run uh cloud in such a way that it can access your local files on your computer and help to build stuff for you uh locally. So it has a lot more functionality than it can otherwise have via the the main uh interface where it only like responds in text or certain widgets that it supports. Okay. So uh it also has the option to connect a bunch of other I think different plugins or connect to different like repositories online and things like that. So like basically you it has more options than what the normal cloud uh provides. So I'll give you an example of something that I built in cloud code. uh to to show you what what is different. Uh I'm gonna go ahead and share. So um the ID that I'm using in this case is Visual Studio Code. Um it's not necessary. You can use your uh like you can even just run it from terminal or uh like if you're on on Mac or I I believe there are other like like just run it from I think command prompt in uh Windows is the other way that you can uh run cloud code from and basically it like claude will have cloud code will have access to like um you can set up like project folders with on your on your desktop and it can create like web pages for you and different apps for uh rather than just um limited to the normal just cloud screen uh in the desktop interface. So in this case I asked it to um I want to showcase a few things. One is something that I have oh no I want to showcase in this uh for for this use case is skills. So if you have uh a task or job or some some prompt that you've um asked AI to help you do and you need to do it regularly uh then you should save it as a skill. So um in this case I have created this skill called find backlink opportunities and what I've asked it to do is basically do a few tasks. So um you can see how it is going through the logic here of the different tasks. The task I asked it to do is um I asked it to find uh broken back links. I also asked uh AI to find uh best pages based on the number of links that they have filtered out for spammy links. So I want um pages that have high quality links linking to them on my website. And then I asked it to find links of my competitors. Um basically link backlinks that my competitors um have but I don't have and do that all into create a summary report on how which of these uh back links are worth targeting. So that here it shows uh that it has everything it needs and it will compile a full analysis. So then it analyzed um for this is the website that I entered in. I said, "Please find backlink opportunities for this particular uh website." This is our help center, by the way, not our main website, but I just use it as an example. Um, it showed me the top three competitors that found for that website and uh it helped to look at uh broken back links. So, again, this is um data from HF's uh MCP and um it showed basically suggestions on how to to fix. Sometimes it's redirects and sometimes it's recreate. Okay. Um this skill I will um provide as a way to that you can add to your cloud code um like uh after this webinar ends. I just want to show you what is uh possible here. But it's what I did earlier except for backlinks and then I saved it as a skill so that I can call it again uh later on. Uh so then it gave a list of like broken back links. Uh and then here it looked at linkbait content ideas. So I asked it to look at the the top performing uh pages based on links pointing to it. Uh in in the ATS platform, this is the best buy links report. Uh and I asked it to group it into like different categories of like what kind of content basically attracts links, right? And then um there you go. It just gave it even suggested like content ideas that I can start writing. Okay, here as one of the steps uh and then here um we have competitor uh link gap. So this is the step I suggest earlier where like I want them to look at my competitors and see what's ranking for them and then um basically find uh page ideas but also places to reach out here like top 10 pages to target for outreach uh as um to ask to build backlinks with them. So it gave also some suggestions for doing uh link outreach. Okay. Uh there's a lot of people writing in the chat. Uh yeah, so um it's very easy to um set up uh I'm sorry I'm I've not used I've personally not used co-work before. Um so those people who that have actually used coowork they can probably uh answer this question. But in this case with claude code it's very um because uh claude code has a function to help you build skills. So um in order to you can literally ask it to say okay please build this as a skill and then give it like a name. So in this case I have given a name of like uh find backlink opportunities. I said I want to make this a skill. Uh and then it helped to create a skill based on that. So that's the magic of AI. You don't even need to to know the specifics of like oh what like uh formats I need to give it necessarily. A, you can ask AI to tell you that and B, based on like what you write in English, you can um ask it to help you to build functionalities that you can then improve your AI usage in later uh times. Um, okay. So, like it just gave me a bunch of instructions here. I can later on basically ask it to um build an app for me or uh create an additional like report format for me to present it to if I have clients or if I have teammates that want to see this in more detail. But you can see all the information here provided is based on my complex uh instructions for it to ask um AI to provide uh opportunities for backlinks based on a particular uh website here. Yeah. And propose a plan. I asked it to propose a plan as well like which would be the the um things that I should do first to build back links then uh later on once I've done that what I should do and then finally um how I should like the you know progressively increase the number of backlinks linking to my website. Okay. Um let me just quickly go back to my slides to to finish talking about this thing about skills. so what are skills right? So I mentioned earlier about what what skills are. But um in order for me to under like build a skill, I asked Claude or you can ask your AI what is a skill and how do I set one up? And it actually provided me uh with this uh write up to say okay you should tell AI or cloud code in this case where you want the skill to be saved what is the name of the file you should give information about this name and the description and then I put all my prompt descriptions here about what I want uh this skill to do for me. Okay. So I can copy it here but it would be easier if you just ask your AI that I want to create a skill. what format should I give uh things to you so that you can make the skill the best right so and then afterwards as you saw you can just call it again uh like with find backlink opportunities and in fact um in the skill itself you can give instructions for for uh to to AI to say use this skill when I say x or when I'm trying to do x task so sometimes you may want to use this skill without even mentioning the skill itself uh Mary uh I will um I have not done it um yet because I needed to I wanted to refine the skill a little bit before sharing it with you guys. But at the end of this webinar I will send you um basically a way to install all these different skills that I uh presented in today. Yeah. U but don't use this as a l to not try building a skill yourself. You should give it a try. Um, not only should you try to make your own skill, you should also try to get AI to teach you how to make that skill and you should also get AI to help you improve that skill. So, for example, uh, in the find backlink opportunities report that I um like skill that I showcased just now um it was using the wrong API calls. So um in the progress of running the skill AI itself realized that it was not uh finding it was not succeeding in calling the APIs uh for MCP I mean like calling the correct uh HFS MCP connector properly uh and it fixed it but I asked it to say okay now that you found the errors please update the skill so that the that that the um skill like calls HF's NCP more correctly and it basically um AI figured out like a schema and added that to the skill to say okay when you run this skill use these exact you know um like a HFS MCP uh connector uh calls and stuff like that. So that's something also you can ask AI to do for you but definitely if you give that a try um make sure that if you see that something's not working you can ask AI to fix it as well. Okay. Um all right app building time. So this is something that um I tried to spin up to see whether or not uh I can build something. So why why build an app? Uh the benefit of building an app is that if I want to have something that's interactable. So I don't want necessarily just like a long page of reports of things. I want to be able to see the the data in a more dynamic fashion. I want to be able to look at particular charts or maybe I want to do it use the data or like the analysis that I want to do in a more uh regular fashion rather than and through through a UI rather than calling like AI to keep doing this every time. So it's good for there are a few way reasons why this might be good and and the like an example is that if let's say you have to do the same job over and over or like you have to just upload a similar file like regularly and you want things to be very consistent then uh then you just ask AI to say okay I don't want you to keep using AI every time to to like do this analysis for me build an app and then I can uh run once you know the app is working I can basically simply do this analysis myself. So, uh I will show you what I made here. So, um this is just a server that is running locally on my machine. Um, there are ways to be able to if you use like GitHub or some uh like Git online, if you have like kind of a Git repository that you have online, you can ask AI to host it as one of your online repositories and then redirect it to like a URL uh in in on the internet and uh like like a domain or something and then and based on that people will be able to access what you Uh so but it has to be hosted on a remote server not on your laptop. Uh that's what I from what I understand that's like the most common use case of how to make it so that you're uh like if you want other people to access what you've built. I'm sure that I'm glossing over a lot of different ways that this can be done otherwise, but this is like the difference between like here I'm running it locally on my machine, but like no one can access it uh because it's not like hosted somewhere uh on the internet. It's like they're running everything from like some folder in my computer. Okay. So, uh in this case, I've asked AI to basically build a report where I want to look at different trends. So, I want to see if I can uh show you what I've asked AI to do. So, let me pull again pull up my Visual Studio Code. So, um it should be this one. Uh so at the very beginning okay uh I think this is not the terminal that I have on or I think I cannot see what I posted earlier but basically I asked it to look for let me see if I can give an idea I'll go back to this screen to explain to you what I did. Okay. Uh but first let me go back to sharing my the app that I've created. So what I asked you to do is basically look at like uh 20 keywords that were found in the last six uh last six months related to a particular uh like for a particular website here and I want to see things like and I want it to grade based on what is the uh score of how um its potential for building new content around. So, uh, I've where is the original prompt? Um, I've asked it I I'll post your original prompt. So, basically, I said I want to look at 20 uh terms that are found in the last six months that have um that cover beyond what I currently cover for this particular website is what I said. Uh, and then I asked it to provide like useful metrics. And if you if I click onto it, I can also see I also asked it to can you please pull the SER for this particular keyword so I can see so I can see the top ranking uh pages and then do an analysis um on on the on these top ranking um results to see like how um what would be my strategy in order to also rank for this keyword. Okay. Um and it's giving information like based on the different SERs that are available you should consider partic different kinds of content structures. Okay. Uh and yeah it then grades it to say like okay it will grade it based on things like how difficult is the keyword, what is the search volume, what is the opportunity to be able to rank for it and stuff like that. And then I've provided a way to create different like filters and in order to see uh information based on different like uh dimensions here. This was built in cloud code. I did not really give it any specific information about like how the UI should look or um how you know um what the formats of these pages should look like. I just said I like I described what I wanted to do based on function. Okay. Um and also I've like asked it to uh so let's give I just want to go back into Visual Studio Code to show you the process of how I went about building this dashboard uh because it was not correct on the first trial. So that's what I wanted to show in the screen. I wanted to show the original prompt but I think I have to go back in history and I can't see it. So um I found some errors when I was running this project like um I then copied and paste the error message. So okay, what I did first is I said um to claude code after like putting in a prompt, one helpful tip would be to say make um appropriate error codes at every step of the uh process when doing the analysis so that we can trace where an error occurs. So um and it did that it create basically whenever they it found any issues of like running the analysis or or running the dashboard it would push an error that came out and then I just copied and paste the error into AI and said can you please check why there's a problem and then uh then it said okay so I found the issue like the API key was not entered correctly in the correct place and then it bas and then the difference here is that it was adding new lines of code and deleting the old lines of code and so it show you the whole process of how it goes about fixing your your platform. So then it's like an iterative process of trying to tell um AI and and testing the workflow to see whether or not something can be uh fixed whenever an error shows up. Right? So then it showed me a diagnosis to say okay this was the problem and this is how we fixed it. So then I restarted the server and then again I I went to check and I said I got this error. Then I just copy and paste the error in and say like there are some issues with the API. Um and during the process of of it uh analyzing it found that it was using old documentation somehow or like hallucinated documentation for um pulling the uh API correctly. And so um basically I went in there to go and check um it um it would show you like information about what what it's trying to do and I helped to put in like the correct URL that it should use based on our documentation. Okay. So like this is a back and forth place of like trying to you will have to go through this process to be able to fix it. But I found it to work really well just by pasting in like um information about what errors that I experienced like what what kind of problems that I faced or or like even if there's no error message I would say things like I have you know requested new data to come from HRF's API but I don't see anything on the dashboard and based on that it will try to debug. can you share what all the app does? uh in this case I think I shared earlier but uh it has a summary of like um all the keywords that analyzed at the top here and for every uh trending keyword that um this website does not rank for right so it's looking for uh keywords other than what this website ranks for uh that has like at least some minimum search volume and I asked it to also pull the SER Uh and then it shows that AI overview showed up as well as all these like top ranking websites and then it does some analysis. Right? So based on that I can basically see like very quickly if I want to um uh rank for any of these keywords I can go into more detail to figure out if I want to target it, what sort of uh content I should create and then copy I mean compare the different keywords side by side of what is worth my time. uh like uh trying to build content for uh okay, Kai, can you there's a lot of uh questions in the chat and a lot of questions Q&A. I will answer these at the end. So, let's make sure we move ahead because the hour is almost over. We've not covered all the stuff yet. So, I will try to go back to our slides so that we can follow along. Where are we? No, it's the correct one. Okay, let's go back to my slides. So, what's the next one? Okay, so now I'm going to showcase. So, somebody asked earlier, I saw that like how can I share this with my teammates, right? in cloud code natively like if you just just connect the HSM MCP and you want to to spin out something and if you don't have like uh like uh fundamental knowledge in uh coding to be able to learn how know how to run a server that people can access or manage authentication codes and stuff like that um it can be difficult to share with people. Um there is another platform that um that you saw in our documentation earlier that we say we support and that is lovable. So lovable has a uh the difference here is that I think they make it very easy for you to share with other people uh what what apps that you have built. So in this like publish thing here I've not published but I can publish um this application that I have built um and then I can restrict it so that uh let's I can I can restrict it so that only people in my lovable workspace can access these um this application. So this is an example of like what is the difference between like lovable and cloud code. One of them is just the the how easy it is to be able to uh publish this application either to the public or to um like your your co-workers or something so that they can also use like an application that you've built. Okay. So um here I have I'm able to show the prompts. So I basically shared that I want to create the app that helps me track the position of HFS over time in AI generated responses in multiple platforms. and I I asked it to give some aspects of what to analyze and for each of the aspects it basically created a different tab. Okay. Uh here you go. Um so so like and so um I want to see like how it's its citations are are like how the sentiments are in different categories and how they are pulled over time. I want to be able to manually uh for example I want to come back to this application in a week's time and then do another poll of all this uh what AI is saying like the most recent data on AI responses and basically get an idea of how the sentiment has changed. So so uh and it tried to do that. So in this case in in uh lovable what they do is that based on your prompt they will create a plan of what to build and then it it will say things like okay how to store the data how to like what are the features and how it should look like it should have this page this page this page so this is something that um AI would basically provide uh and then based on that um it will say can I proceed with building this it'll say plan approved and it start to build okay but here's caveat about lovable um a lot of times they will build something and if you don't give specific instructions to test what you have built so whenever like for example in this prompt here I didn't give instructions that like after you build this application please test to make sure things work and that was my mistake because I then went through multiple processes uh multiple like iterations telling AI that like hey I ran something and there's no data or I ran something and there's no analysis created so please check. Uh so it it um that is the so I ask questions like why is there no responses and then basically tries to investigate the issue. So it is so useful when you um prompt to AI that you keep in mind that AI should be able to test for you if what they built actually works. Okay. bonus is if you're trying to build something and if you already have an example of something that is working like a example report of what exactly you want to see you can attach it usually into AI to show uh in lovable for example I can like make a screenshot like if I make a screenshot here and if I want to paste into lovable and say like I I see an error um uh in this screen please investigate that's something that I can also uh depending on your AI platform they might support this functionality for for building apps. So like so definitely keep that in mind. Uh but yeah in this case I I basically asked it to say can you please look at the AI responses uh about a particular brand. So I asked it to set up a brand and as well as the website. And as if people of you that are familiar uh with uh what this looks like is let me show you uh what I did here. I because like it was uh this particular app had a lot of issues. So I asked it to keep a log of all the APIs that it pulled. So uh in this case it said yeah I I pulled API from uh mentioned. So these endpoints are mainly in brand radar. Okay. Uh this does mean that if you want to use HSN with brand radar, you need uh the add-on. For example, in this case for AI overviews, you need the AI overviews and AI mode add-on. Uh you can also create this similar dashboard um with of your tool of choice uh with custom prompts which is a lot more affordable. It starts at like $50 US a month for for some prompts. So, like let's say if you don't want to um purchase the entire brand radar add-on, you can just purchase custom prompts and then fill up the custom prompts uh on hrefs and build a dashboard for this different like in in just to analyze your custom prompts and that's a lot more affordable. Okay. But yeah, I asked it to basically look at all the different uh um AI responses across different platforms here like AI overviews, AI mode and chat GBT. And I asked it to give some examples of like what people um what are the what AI has said in the responses that are positive and negative and uh and unusual. Okay. And then I can also um because I've asked for it, I have asked the an ability to see the actual um AI response as well here. So I can basically look into it and and see more information about how it works. Okay. And and like the whole purpose of this app that I tried to build is to be able to see the changes of this over time. So then I can change the date uh to like next week or the week after and basically be able to see how that changes over time. and then narrow down on the changes. So that's something that you can ask AI to do as well. Okay, I hope that was a cool example. Um, but just note that like with app builders like this, there are some like limitations that you have to uh keep in mind. Uh, especially and it's also like experience. The more you try to use AI to build stuff, the better you understand like okay, these are the things that I should tell AI ahead of time so that I can avoid the same mistakes. Okay, so we are at um like an hour done and there were a lot of questions that people asked. So I just want to quickly go back to the presentation and just go through what I wanted to show u very quickly and then go into the questions. Okay, the last one I wanted to show is the um automation of uh with skills. So, um I will not show you the skills themselves because um I've asked our teammate Ryan Law to about that and he's not comfortable sharing it. But I will share with you um his blog on it. So he goes into detail about how to build it, what he built, and how you can build it yourself. So, I'm going to link this here as well in the chat where basically um he did um a whole workflow of being able to say prompt create a starting prompt, add an um you know, create a content brief based on your starting prompt and then create an outline, then edit the outline, then add links, make sure you mention like HFS's products and features. So he has like a whole outline here and the success of what he has built is primarily because of how well he understands the blog creation process and you can like read this whole um blog to go into detail as to how you can do this. But like with skills you can actually so here I've mentioned that you can get a skill and tell the skill to prompt multiple skills in succession to get basically and chain a bunch of different processes together. And that's basically what he did in this process. So he could automate a lot of his like content creation process and then just focus on creating like examples and templates that AI can use and generate so much more blogs and content than he than he would otherwise use. And he in this blog like when people are asking him of like examples of how to prompt AI this particular blog goes into detail on how to do that. Um I can very briefly I think I hopefully I have saved it. I have not. Okay. Um I think I want to go into blog pipeline. Can I do research socks? No. Okay. Um sorry. I just wanted to test something for a minute. But uh basically this is what you should check out if you want to automate like a full-blown process of creating content. You can you can see how he did that here and this should give you inspiration how you can do this for yourself and how so each one of these like different process here is a skill and then he had one master skill that said please uh execute all these different skills in succession. Okay. okay. So tips for getting the most out of HF's MCP is you like I mentioned earlier ask AI for help. If you face an error or something is unexpected just ask AI. I said I found something um that I didn't expect it like the the app that I created or the analysis that you gave me did not match my expectations. Here's an example. uh and AI will basically give you ideas on um it would investigate the issue and a lot of times it would help to fix it as well. Um if you want to use HSM MCP especially for certain uh like if it's not cloud code or maybe a less supported uh AI platform you most of the time you need to be very explicit to say use hrefs because sometimes if you just say can I do like a backlink analysis or maybe you've connected with another uh like SEO tool platform or something like that and uh it doesn't know which connector to use you can be explicit to say use hfs to find something or use hfs to build this app uh and Yeah, the more details you can give uh as well as more examples of what to do or not to do when you prompt AI um make sure that you get a better result uh like you know with every iteration of asking talking to AI uh and um there is a final part to that I mentioned earlier that if you're able to tell AI please test what I've built and while in that original prompt it would basically be able to build that in a workflow especially especially if you're building either a skill or an app to say now that you have created the skill or app please test to make sure that the results are are exactly what I'm looking for. Okay, so these are all ways that you can help to reduce a lot of headaches later on when using HSMCP. and here are some resources. Again, I will share this deck with you guys later on uh at at the end of this presentation. But like here I've I've talked about the documentation. Um there is uh Glenn's 15 NCP use cases here which are mainly similar to examples one and two that I shared. Um the app um related ones I have basically um created outside of this use case list. But um I would still def recommend you check out his uh blog because it gives a lot of ideas of what exactly you can call HF's MCP to do. Um Tim our CMO actually published a public repository of some skills that he has built. So um but after today I will provide um a repository of the different skills that I mentioned in today's webinar but if you want to check out what he has built for finding new pages you can check out this skill and of course you need the HSMCP to uh to use it. So you can um get go to this link. In fact, maybe I can copy it for you guys and and then you can go to it now. Uh and you can see how you can add that skill to your um AI of choice now and check it out and see how it works. And uh there's also this one um page here where the community has come together to show some best practices for building skills because we've met um showcased some of it today and so that you can create it so that it works exactly as how you want and uh be able to avoid a lot of problems that are commonly used when creating skills for the first time. Um also um for those that have stayed around, I know some of you guys have left already but that's okay. Um there is something hidden in plain sight on our front page and I don't want to spoil it but you can now all of you guys go to our um atres.com and scroll down to the footer and why did I mention it today is because uh remember I shared uh about this app that I built in lovable using the HSMCP. Um there we might be building something related soon that we're excit that that soon that we will share. Sorry where's my slides? Uh, and for you to get early access or to know about it, for know more information about it, you have to check out something that's something interesting on our front page. Okay. All right. So, that's that's that's all I'm going to say. All right. So, we are now at the end of today's presentation. let's go with the questions. So, here's what I'm going to do. I'm going to look at the Q&A section. And then uh I will quickly see uh in the chat um if there are any questions that I did not get to answer in Q&A. Okay. Uh let's do by most up votes. Um Arthur asked, "Will you get into how much API units a typical MCP request costs?" so there's two layers of this question. there is how much API units does a request cost via our HF's API and you can actually see this when you open up a report. So let's say I'm going to go to site explorer because we have a lot of API endpoints there and I want to let's say pull out data from the overview. You see this like API button here on the top? So you can see the request like what is the API key that you need to call uh as well as the cost. And the cool thing is if I let's say go into this and I apply a bunch of different filters like let's say I want to look at lowhanging uh fruit keywords. Uh can I already apply it? So if I apply filters uh to my report, maybe I only want to see United States, right? Then when I press this API button here, you can see like it would update. So basically um the API request will update based on what filters I've already have selected and it will also show you the cost per row. And then of course I can um dictate how many like rows of results I want to look at. So earlier apps you saw that I only wanted to see like 20 uh keywords or I only wanted to see 200 results, right? So that's how I like cap and that's like gives me a rough idea of how much it should cost. Um however AI may choose to like call the API in all kinds of ways. So like that's sort of the second layer. it may call the API multiple times. It may call um the the wrong API or it may like and then you would basically waste a lot of uh API uh units going into some data that you can't use or stuff like that. So and that's because AI is not very consistent unless you give it like explicit uh instructions to uh so the more explicit your instructions the better you can control the API usage via MC MCP server. So that that is sort of the limitation of AI. And so like my only uh feed suggestion on how to help manage that is to basically track this with your API keys and like because I have them for different projects or or different connections, I have a rough idea of how much is getting spent and that helps me um give a track of like how much units get get spent. So it really depends. There's um and it also depends on like how complex your requests are via HSN, NCP. Are you calling, you know, just for one simple task or you're trying to chain multiple analysis together? Okay, I hope that helps to answer the question about units being spent. Um Nick Trionov asks, is there a way to use the LM of choice to track our API usage or is there a metric that isn't shared over MCP? okay. Now, docs.hfs.com or maybe I can showcase this via Let me just check one thing very quickly. So by default when an HF's API is called it gives the unit spent uh in the API return request and there should be a way to do it in our documentation our AI document API documentation we do give uh information about how to do that so I think I'll share this particular page. I think you can request AI to to give the uh unit cost because by default with the API for example, not by default, you have to make sure that you call it that particular function to be able to see the cost and the output should be able to give you like the units that were spent basically. But the I think so so like you can try to ask um AI because in the API that we provide you should be able to see the the units spent per um for for the request. Okay. And if it's not able to show that information for whatever reason, you you can try to contact to our support, but like we you should be able to see the the units like and like if you call the HF's API normally. So if AI is not doing it for whatever reason, maybe there's something with your prompting that can be improved. Uh Agus Hermaan asks, is the site audit/technical data also connected to HSNCP? Yes, it is because we have the API endpoint for it here. This one while we're on this page. So, here we have these are the different site audit uh APIs that we support the score as well as the issues and it give you basically a JSON of like what are the different issues and number of issues across different categories that are there. Uh you can even get the page content itself. So yeah, so um yes, uh site audit is supported, but like that's the that's the overall answer, but if you're looking for something specific, a site audit that is supported, not supported, you can check our documentation to see. And as long as you have that, you also have to make sure that the MCP uh key that you use is in the same workplace workspace as the project that you have run your site audit for. Okay. And then obvious obviously make sure that you prompt the correct project name and things like that. Uh but yeah, if we have it in our API, we should have it in MCP. Which HS plans is MCP available on? Light and above. Now same with our API directly. You can access it from light and above. Uh but the limits differ. So, so the limits for for all the different plans that you see here in integrations. So, there are two kinds of limits. The first is the number of units per month that you can access across either direct API or MCP HSNCP and the number of rows per request. So sometimes like if you want to do um if you want to request for more than 10 results per API call you have to upgrade your plan. So those are the sort of like limitations. So higher plans obviously can basically pull a lot more information at once but uh yeah so hope that answers that question. Um, Jeremy asks, Jeremy Kaiser asks, "Group categories, not the parent category of atres, right?" I think you're referring to that use case where I grouped a bunch of like keywords together or I grouped it I grouped it based on uh different backlinks or something there or sentiment. Yeah, that's not um hfs not something that is done in hrefs. I asked AI to find like a best fit method like uh either you can use some data data science method for grouping like semantic keywords together or you use AI to do that job and they have different pros and cons but it's not part of uh atres you just ask AI to sort of um propose to you ways to group your data together so that you can digest it better and it usually does a pretty good job. Sarah Bodto asks finding backlinks use case is doable with just chat not clawed code. Um you can do it with like chat desktop but uh I think you can even do the skills um in and and create the skills in cloud like cloud desktop these days or even a cloud chat. Uh I have mainly used cloud code to create skills because I find it easier to manage if I have different projects and and different folders. I can uh organize that better like I don't want all the skills be messed up. So I I did it in cloud code specifically but you should be able to do it in uh like the normal cloud. So I saw a way to do it in a platform the other day. Um I think it's under but there's skills. I know that there's skills um in in the normal chat. Not sure. Sorry. But there but there is a way to do it. I I can't find it now. I saw it. I saw it before, but yeah. Uh hope that helps somewhat. Um, okay. So, Jennov asks, there should be a way. I I'm not able to find it live here. I also just updated my claude and they added some stuff and I have to figure out what they claude and cloud code updates very often. They keep adding new stuff and I have to keep learning the platform every time. Uh, anyway, Jenna Van Hull, sorry if I butcher your name, asks, "Do you have skilled templates for common tasks?" uh in Hrefs we will be providing that soon but uh I think if if it's to do with HFS NCP we we don't really have something like that you can download and access just yet but we're working on something like that. Uh in the meantime I would recommend you go back to the 15 use cases. There you go. this one. Oops. One second These for now. This is all I can share with you. Okay. For common tasks. Um, and there's 15 of them because there's five very simple ones. Uh, and you can copy them. You can grab the full version and copy and paste all these and create basically use that to create skills. Uh there's some that are very simple and it wants you to do like just like pull one or two API uh calls and some that like have do multiple tasks and and like at the bottom it basically does like a full full-blown like analysis based on things that you want to do. So like based on that you can create these skills for yourself to use and reuse. uh when you run the backlink analysis, what do you do with your information? So, how do you outreach? So, Claude will not be able to help you do the the the link outreach if I understand your question correctly. If I want to do link outreach, I have to find the owners of the pages that Claude has recommended to me that to say, hey, you know, getting your brand mentioned on that particular on these pages, these 10 pages would be a good idea. You can reach out to the owners of these pages to see uh whether you can work out like you know some arrangement or build a relationship with them to then get your brand mentioned either on this page or maybe future pages. So yeah, uh Claude itself won't be able to help do the outreach, but there was a section that I look for which is broken back links. That's something I can control myself. I can look at the pages that uh lots of uh people link to that are actually broken on my website and then I either set up redirects or I create like new versions of the pages that were previously broken that have updated information. That's also something that I can fix and basically I reclaim all those broken back links. So that's something I can control. Okay. Um, is web analytics available via MCP? Sorry, I don't know the question answer to that just yet because yes, it is. And I think this is very new, right? Are people able to see my screen with this? Oh, uh, I have to share a different screen. There you go. Uh, is web analytics available via MCP? Yes, it is. I think this is very recently available. So, definitely go check it out. If you have it set up on your uh HF's dashboard, you can check you can try to access it and create um dashboards or apps of your own. Uh, Sana asked, "I've experienced HS MCB pulling wrong data points into my workflow. Is there any suggestion to counter data hallucination? this is kind of an ongoing problem with AI in general. I think it used to be pretty bad and it's gotten better. Uh, I I find it very it's not something that I can guarantee you that you'll not get hallucination in the future, but you may ask AI when creating the like pulling the data to be able to put some guard rail. So basically say like if there are certain data that you pull that is let's say I want to do keyword research like the first example that I shared and um I can ask AI to and when they pull the answers to detect keywords that do not seem to semantically match what I search for and then basically identify those and ignore them. So that's something that you can build into the workflow. But there's sort of not really a reliable way necessarily to block hallucination cuz that's sort of integrated like inherent with AI um specifically. I experienced it across every tool that I presented today. I've experienced in loable and cloud code. I've experienced mistro. So, I don't really have a good answer on that other than what I've proposed just now, which is to try and put some guardrails to like check to see if like certain things don't look right and then in that process of pulling the HSNCP data, try to ignore those examples. Um, okay. So, these are the questions in the Q&A. I'll just quickly look through here to make sure that I answer people's questions. those that have hung around and still have any questions. Now is the time to ask. you see that overall people are like using claude for sure. Um some people have asked what is the difference between uh HF's claude connector and HSN MCP. It's so um when you see in claude that HFS like the HFS connector NCP connector is there. We've just officially put it on the platform, but a lot of um thirdparty providers out there provide an MCP connector that's just not available as in the default like marketplace. You can add them via like add custom MCP. And functionally, it's the same. It's just like presenting it in like one official way versus like uh versus like a custom uh NCP URL. So, it's just uh like it's whether or not you're officially in their marketplace or not. are there any plans for adding it to starter? Currently, we don't have any plans to add HSNCP to starter. Um it would be available like plan and higher. someone asked to target uh Julia asked to target non-English-speaking country. Is it better to prompt in English? Uh, prompting in English just helps you. I think prompting an AI in English which country is targeted or is it better in your prompt in other tong? So what I understand based on what my colleagues tell me is that English if you prompt AI in a language other than English it does not necessarily work too well but it depends on a language. So any languages that you see AI support relatively well, I think they would similarly translate to how well they can understand your instructions and build things in AI. And if you're trying to do research uh and ask HFS's NCP to pull data uh like for you to to do that research, I think you can for example upload a CSV file with a bunch of all those like non-English terms especially if it's like uh nonromantic languages like Chinese or Japanese, you can put those um words in there and then upload it to AI and then ask them to say okay do analysis of that and that helps to sort or contain things. Uh yeah. Okay. So I think some of these questions we've answered before and yeah like whatever I've presented today I will um there will be a follow-up email when when this webinar ends uh to share with you guys like a way to download like everything that you saw today and try to run those skills uh for yourself to see if you can build them yourselves. But of course definitely try to do what I did today to um like build stuff in an iterative process because that gives you an idea of like what exactly you should like the whole point of like AI and MCP server is to create something that you need and and what I'll be providing is just Uh okay Venet Jane asked can skills be used via cloud API. Yes, it should be possible the same way that it should be possible in cloud code. Yeah, Vishek has been here helping to answer questions for people. Thank you for doing that. [laughter] it's been quite useful. Uh, but yeah, do we need to create NCPs of our own websites for better results? This is seems seems to be an M a like a AEO question, I guess. Uh's answer here is that not necessarily. So I think charlam sharp sorry Charla Mapos you've asked if we need to create MCPS of our own servers uh with the idea that maybe MCPs of your servers will allow you to do uh connect like people to discover your website via AI rather than through search. Uh it's a good question. I think a lot of SAS providers are doing that as a way to get to being getting discovered. Um, but and and like platforms like, you know, as you know, I think trip.com and all these big companies that offer a way for you to like buy flights via AI or something like that. They're working on things like this. So, the future may be more tending towards this like NCP API environment where a company just works entirely without a website and through MCPS, but we're not really there yet. Um there's a lot of advantages of having a website. I think like as much as you we all rely on AI to do certain things. If I want to know a brand, understand a brand, sometimes I still have to go to that website directly and learn about learn about them or or check what they have or maybe learn about customer use cases and things like that. So, um, and in the other hand, like there's advantages of having both like MCP and a website, uh, if if like depending on like what your business does. So, um, I guess it's just a matter of whether or not you can create something that's useful and unique. So, yeah. Uh, yeah, I I'm not sure if I can really answer this this question confidently about whether or not Oh, you mean for doing analysis of your website? Okay. Um, as opposed to running the analysis through site audit and pulling that uh via HSMCP, that's something you can do. I I think that they are I'm not sure if it's better. Uh I know people have tried to do something along those lines and I know that for example um certain uh like Cloudflare is providing an API that you can just like with one API call like download an entire website or something. So uh I think there are ways to go around things without necessarily building an MCP. So I think lots of people are trying to build ways of um trying to analyze a website at scale. So I think you have a lot of options today but I can't tell you like which one of these options are better but I know people are building them. Um yeah uh Ari Vita asked the question of like developer mode and GP GPD disabling memories. Uh when will MCP be available in regular mode? Uh I did try I did look into um like show showcasing chat GPT today as an example but as as you've rightly pointed out it needs like specific privileges that like I did I could not set up in time. Um I can't answer questions on behalf of chat GPT. I think this is a matter of what OpenAI wants to do to be able to uh allow more NCP connectors to work the way people expect them to work. Which is why like you see a lot of people in today's webinar are using claude. So a lot of people have decided to use cloud instead or I saw a Gemini and stuff like that. yeah, I don't know. I can't speak on behalf of open AP uh open AI. Okay, so we are half an hour past the main hour for our webinar today. I think I've answered most questions. If you have other questions that you like to ask, there will be a follow-up email where you can ask like email them to me. I can try to answer them best I can. Um especially if you have any suggestions on how we can like improve HFSMCP for you or like certain like use cases that you said either you tried to build or you want to build you want more guidance and stuff I can definitely help you answer them over email. Um but for for today I will um go ahead and close the webinar. Um, when you close your Zoom, you'll be prompted to fill out a survey form. And that survey form is just our way of getting uh information about how you found today's webinar, whether you found it useful and and things that we can do to improve. Um, there were a bit of hiccups today which I like I'm a bit sad about, but overall I think I was able to help showcase different ways to use HSN and certain like tips on how to better use it and like showcase what all things are possible um with AI and HSNCP. So I hope you found today useful uh and um we will go ahead and see you guys in the next webinar. Oh, quick shout out quick shout out to next week's webinar which I hope you will not miss is this one uh with uh Patrick versus uh Ryan Law. So you saw today that I mentioned about Ryan Law's AI workflow that he has built entirely by chaining different like skills together. Um, next week we will have a showdown where we will compare where they will fight with different HFS use cases against one another and you guys, the audience will get to vote on which ones you found useful and the winner will be crowned. Okay, so um the previous um showdown happened last month with Glenn Elop and Tim Solo and people absolutely loved that webinar. So you do not want to miss it this time because we have limited seats in our webinar and like once that runs out basically you will not be able to like we have max 500 to join. So if you don't uh make it in in time you don't make it in okay and you only get to you don't get a vote basically or or like but you can see the replay. Okay. All right guys that's it for me today. Uh thank you for joining us today and I'll see you guys in the next one. Uh take care and have a good weekend.