Build Anything with MCP Agents… Here’s How
let's talk about mCP or the model context protocol this has been blowing up recently a ton of people are covering it especially on YouTube but unfortunately most people talking about this have no idea what it actually is how it works and how to explain it in a way where they're not going to confuse the heck out of you so in today's video I'm going to break down what it is in very simple terms coming from someone who is actually a developer and then I'm going to show you how you can set it up inside of cursor and start using it to build some very interesting AI agents so with that said let's get into it and learn about the model context protocol so in order for us to truly understand mCP servers I want to make a comparison to something that we might be more familiar with which is a rest API now even if you're not a developer don't worry I'm going to explain what this is but the point is that we have these things called rest apis rest stands for representational State transfer and these apis are used typically for web app communication they can be used for other things as well but the example you'll be most familiar with is a standard web application now when we have a web application we have something like a client or a front end the front end is what you interact with as the user something like facebook.com or instagram.com where you're actually you know submitting a form or making a post or viewing comments or something along those lines now when you're using that client behind the scenes what's happening is it's sending these HTTP request using something called a rest API so we have this server the server is secure it's being ran by some kind of company and it has all of the data for the web application so when you make a new post for example you send an HTTP post request to a particular URL on This Server it then makes that new post updates the database and then sends you back all of that data to the client to tell you hey this post was successful same thing if you're fetching comments or retrieving a profile or something along those lines behind the scenes you send this HTTP request using this API and then you get the data back now this is a communication protocol it's standardized it's something that's used probably tens of millions of times across the internet and there are literally billions of HTTP requests being sent every single day now the reason why we have this is to set a standard a consistent way of doing something one where all developers can just follow this practice and we can have all of these millions of apis that work in the exact same way now that leads us over to mCP servers and why they were created and how these can help you build really powerful agents and speaking of gamechanging AI agents today's sponsor samb NOA is pushing the envelope with their open source deep research agents and the fastest most efficient AI inference solution powered by reconfigurable data flow units or rdus samb noa's Hardware stack delivers 10 times the performance of traditional gpus this means you can run intricate research queries and generate insights in record time unlocking discoveries that were previously just complet completely out of reach and the best part is that this is an open stack which means you can bring your own models like llama and deep seek and seamlessly integrate them into the most powerful AI research platform on the market with three times faster processing than top GPU providers Enterprises are not only saving Millions but also gaining the Competitive Edge by tapping into realtime insights across scientific research finance and Enterprise AI imagine a leading biotech firm that could use this deep research agent to sift through terabytes of data and clinical studies uncovering novel biomarks that could expedite drug Discovery with sova's deep research agent these complex cross-domain queries can be ran in real time something that traditional GPU based systems simply cannot match you can upgrade your AI stack today by clicking the link in the description and experiencing the future of deep research with samb NOA so all of that context brings us to the model context protocol now this is very similar to something like a rest API this is just a standardized way for an AI agent to communicate with a set of tools So currently or kind of before this video and before this model context protocol came out we had all these different tools and we know that AI agents are really good when we give them access to these various tools databases Google Drive you know search features whatever but all these tools were implemented completely differently there was no standardized protocol so it made it very difficult for us to integrate 10 20 30 40 tools to our agent because it required custom Integrations custom coding and just a lot of complexity because there was no standardized method you can kind of think of it like this imagine that you had 10 friends that were all experts in different areas but they all spoke a completely different language Portuguese English Mandarin Indonesian you name it whatever they spoke all these different languages it's going to be very difficult for you to communicate with all of these people and you're going to keep constantly changing languages and mixing stuff up because there just no one standard way to do it whereas if they all spoke English it would be very easy they can communicate with each other and you can all talk very simply that's exactly what the mCP is attempting to do bringing this standardized protocol to AI agents so now all of your tool providers are going to write code and write Services based on this protocol so they can more easily be kind of integrated into the AI ecosystem and into your AI agent that's really all it is you don't need to know all of the details of how the protocol works and the communication channels but the basic idea is it's just this standard approach that entropic has come up with and now all of these tool providers are racing to adhere to this protocol so that they can be used inside of tools like cursor for example or inside of Claude desktop or all of these other tools that use this protocol as well so it's super cool you can connect to things like databases GitHub stripe notion memory you know all kinds of tools are coming out every single day and if you do care about the architecture you can read more from the mCP documentation and you can actually write your own mCP servers very easily there is two different transport mechanisms the first is over standard input output which just means essentially running in your local terminal and getting input as a command and output as like text in the console and then the other one is called This Ss transport which is actually doing communication over the network similar to something like an HTTP rest API and you can see that these can run locally or remotely so for you all you really need to know is is my mCP server running on my own computer and is it using the standard input output or is it running remotely for example on someone else's server like entropic server open AI server or a tool provider server and there's a different way to connect to it based on this architecture or this transport if that confuses you don't worry I'm going to show you exactly how to set up mCP servers in literally 30 seconds in just one second but you get the idea okay so that's mCP servers again just this standardized way of communicating now what I want to do is show you how we use MTP server inside of something like cursor to just supercharge our AI agent workflow okay so quick cut here because I forgot to mention this in order for everything I show you in this video to work you need to have cursor installed on your computer and you need to have node.js
installed as well now you can get node.js for free it's just called nodejs.org so just go to that website and go to downloads okay I'll link it in the description and then go here where it says pre-built node.js all right from here select your operating system so Mac Windows Linux okay okay make sure you corre select the correct one so for example if you're on the new chips do the armm 64 if you're on Windows right go to x86 x64 click the installer so the windows installer or on Mac the Mac installer run it and install node.js
once that's installed then these various commands will work if you don't have that you're probably going to get an error so I just opened up cursor here where I'm going to demo how to add an mCP server and then start using it from your cursor agent now this will work inside of something like wind surf as well and it will work in other settings too so just because we're using this incursor doesn't mean that mCP is strictly forbidden to this environment you can use mCP with kind of any type of AI agent framework and as this becomes more mature I think you're going to see a lot more mCP servers and a lot more companies adopting this protocol anyways cursor is one of the first ones to make this work really seamlessly so I want to show you what it looks like and then I'll show you how to install it so you can see that I have three mCP servers that I've uh kind of created here that I've added cursor I have a fir crawl one which allows me to scrape the web I have a coin cap one which allows me to get the current price of cryptocurrencies and then I have the sequential thinking one which allows models to think sequentially through a problem okay a lot of people were adding like GitHub ones and other random ones I wanted to do some more unique ones in the video so that you got an idea of how to install these from ones that were maybe smaller mCP servers or less documented now just to show you an example you can see in my terminal here that I was kind of asking some questions about my Instagram and you can see we're actually using the mCP tools and it will show you like what it's actually calling the mCP tool with so here's the query here's the limit and then it kind of gets a result and then same thing here for the price you can see it said you know get crypto price it gave the P the name of ethereum sory and then we get back all of this data with like the ID the name the supply the current price and it shows me what the price is at the current time so if you have this agent enabled inside of cursor then it will automatically be able to use these mCP servers but let me show you how we add them so I'm just going to go ahead and delete all of these by deleting them I can literally just delete this entire file this mCP Json and I'll show you how we add them so to add them you can go to your cursor settings so if we close this there's a few ways to get there you can go to file you can go to preferences and then cursor settings or you can hit control shift p or command shift p depending on your operating system and you can type cursor and then just click on cursor set okay so open up cursor settings go to mCP depending on your version of cursor this might look a little bit different and then you're going to press this button that says add new Global mCP server now when I press this it's going to open up this mCP Json file however if you don't already have this file created it's going to ask you to create it so just create this Json file and then you'll be good to go now from this Json file this is where we can add our mCP servers and we can just put them in a list it's very simple and I'm going to show you how you can find mCP P servers to then add here so I'm going to bring up Google and what I found is this repository I will link it in the description which is essentially a list of mCP servers it's just a get repo here okay awesome mCP servers and you can scroll through and you can kind of find servers that do something that you want so for example we have art and culture we can generate images or videos we have Cloud platforms connecting with Cloud Player kubernetes we have command line ones we have communication there's all kinds of uh mCP servers I found a few for this video that I will show you directly and then we also have a directory called smithery okay now smithery is probably the most popular one right now for finding mCP servers but I need you to keep in mind that this is very new so a lot of these MPC mCP servers s either don't work or are not really clear on how to install them so I'm going to show you how you install ones that aren't super super popular so you get an idea of what's actually happening and like you know what mCP is and how it's functioning okay so we want to install an mCP server so so I found this fir crawl mCP server and the link for this I'll put in the description on this website called mCP doso again it's kind of the Wild West right now we've got mCP servers all over the place some of them work some of them don't work it's kind of a bit of trial and error now from this page you can kind of read through what you need to do here and the first thing that I found is that I need to create a fir crawl API key so I went to fir crawl okay firecrow dodev I made this API key and then I copied it okay obviously don't link this to anyone after the video I can regenerate this so you guys don't see it okay I now have this API key and then I've seen if I go to content that is kind of giving me some instructions on how to add this mCP server if I scroll down here I see I have some instructions on running on cursor and what it says is that I need to go to mCP servers add new mCP server and then it tells me I need to use this fir craw mCP and this command however that's not what it looks like inside of cursor for me right now I just have this Json file like I don't have this popup in order to add this mCP server so what should I do well if I scroll down a little bit I see that I get a command that looks something like this it says if you're using Windows and running into issues try command SLC set fir craw API key like this okay so I'm going to copy this command and what I'm going to do is go back into cursor and I'm going to show you how to manually add this entry so the way that this works is you write a set of braces you write mCP servers like that okay I'm just going to put this in by default so that you have kind of like a list of mCP servers then inside of the MPC servers you write a name for your server so in my case I'm just going to call this far craw okay then you put a colon then you put another set of braces and then inside of here you have a few different options however you can see it's actually giving me the auto complete because I had this previously what you're going to do is type command colon and then you just paste the mCP command okay so I'm going to paste it like that and this is what it looks like now you'll notice that I have an extra set of quotes so I'm just going to remove these set of quotes because I believe that we don't need those and then we're going to paste our fir craw API key right here that's literally all that you need to do so again I found this from the link in the description I'll show you the command and what's going to happen is cursor will just automatically run this command for you it's actually going to open a new terminal and it's going to start running a server that has this fir craw mCP running okay it's running locally on your own computer and then you're going to be using this API key to communicate with the fir craw servers that will then be able to go and kind of scrape the internet I know for any non-developers this might look a little bit confusing I'm just trying to explain what's going on so that you can use some critical thinking later to install other mCP servers now you're going to see that this starts with command SLC if you're on Windows you need this okay A lot of times it's not going to tell you to add this command SLC but all of your commands need to start with this if you're on Windows if you're not on Windows then you would remove that you could leave the same set thing and have this Command right here but if you're on Windows you need this command SL C so that it's going to run this in a separate terminal what I'm going to do and you can see when I run this it actually opens up a new terminal when I save this is it starts running this process in the background now I'm going to close this for right now I'm going to go and put my fir craw API key here and then I'm going to show you how this works and then we'll add some other mCP servers that will be a bit more simple okay so I added the API key and then I just close that Tab and you can see now when I go back to mCP servers it shows me fir crawl and then it says this is the command and it shows me that it's running now I can refresh this and when I refresh this it will just run this again and when it runs it we should see that we now get a list of tools if the command was being executed successfully now again if you want to copy the exact command on Windows it is this okay again it was you know if we go back here you can see what it looks like right here I'm going to blur out the API key but you guys get the idea and then on Mac remove this command SLC same thing on Linux and quick note if you're getting something like client not connected just close cursor and just reopen it okay so I reloaded it here and you can see now by default it's going to spin up this terminal leave this running don't close it and now you should see that you get connected and you don't have that air that's a common thing that I was experiencing anyways now that we've got that open what we're going to do is make sure we go to agent mode inside of our cursor composer and we can ask it something like you know look up Tech with him on Google and when we do that it should know that it can use this fir craw tool because it access to all of these various tools and it will go ahead and do that now it says no tool match the tall mCP server fir crawl uh okay and then it says there seems to be an issue with the tool let me try a different approach and then it finds this fir craw search okay so I go ahead and press run and then you can see what it calls it with so you know searches Tech with Tim limit 5 gets these URLs and then gives me the result okay so we can just go ahead and use that tool and now you can use that inside of your cursor editor which is kind of insane let me show you how to add a few other MC P servers so let's go here let's get out of fir craw and let's go to Smith and let's add the sequential thinking one okay now it's a little bit easier from smithery again I'll leave this link in the description and what we can do is just press on this cursor tab here and notice that I've also created an account and I've just signed into smith3 because they're forcing you to sign in essentially to be able to get these install commands Okay so this page loaded now and you can see that we have a few different options for installation now you can just run the commands that it shows you right here and it should actually just install itself inside of cursor for you so you can see based on the different version of cursor there's a command that you can run so we have this command this command this one right here or if we want we can do the manual installation like I showed you already where we copy this as Json and just put it inside of the um what do you call it the config file okay now I'm not going to do that because I mean we've kind of already looked at how to manually add this what I'm going to do though show you how we add another more random mCP server so for example I found this m mCP server only has 27 Stars some random guy made this but it does actually work and you can see that it tells us exactly how to get this running he has here the config so it just says okay mongodb I don't know why he named it mongodb but that's what he named it has the command npx then the arguments coin cap MCD this one doesn't require any kind of API key so all I'm going to do is just copy this config again I found this from some awesome mCP servers directory I'll link it in the description I want to go back to mCP servers I'm going to go add Global mCP P server I'm going to go here I'm going to rename this to say coin cap like that okay and then I'm going to close this and it says client closed okay that's fine the reason it's not working is because I'm on Windows and I forgot to add the command command SLC so what I can do is command SLC npx okay copy that you can see now that this is running we have it running in the terminal and if we go back to cursor settings nothing's showing up but we get the tools so we can refresh this okay all good and now you can see that this is running and if I come here I can say you know what is the price of BTC and let's see if it can look it up and give me the answer just wait one second here ideally it should use the tool and there you go we get the price of Bitcoin and we can keep using this tool you know as much as we want and by the way if you don't want to have to press accept on all of these like because it's using the tool you have to keep pressing accept if it uses multiple tools and stuff you and if you want to be risky go into cursor settings you can go into features you can scroll all the way down to YOLO mode which I think is right here yes enable auto run mode okay they've changed the name of it to auto run mode but it used to be called YOLO mode a second ago and you can enable this and when you do that now it will just allow the agent to call any commands or use any of the tools that you have now you can actually add some stuff here so for example you can make sure it doesn't delete any files you can have a command deny list and an allow list and a prompt telling it like what it's allowed to actually be doing I would suggest filling in these details personally I don't run with this on but if you want the agent to just like be able to go and do everything and you don't even need to watch it and it can just call any tool that it wants and do anything at all then of course you can enable auto run mode all right guys that's pretty much going to wrap it up I know there was a lot of content there but I wanted this to be thorough and explain what mCP is how it works how to set it up in cursor and how to get over some of the issues that I personally was facing if you want to see me build my own custom mCP server and do it to tutorial on that then please let me know in the comments down below and I look forward to seeing you in another video [Music]
2025-03-28 13:44
