Projects
![]()
No projects found
Investors
New
DAO Treasuries
DATs
Lend / Borrow APRs
Chains
Categories
Spaces
Unlocking the power of MCPs for AI Agents
Recorded: March 27, 2025 Duration: 1:01:11Space Recording
Short Summary
In a recent workshop, Metis discussed the integration of AI agent frameworks with blockchain technology, highlighting the launch of their new chain, Hyperion, and a strategic partnership with NIVI to explore Model Context Protocols (MCPs). The conversation emphasized emerging trends in the crypto space, showcasing growth opportunities for developers and the community's commitment to education and innovation.
Full Transcription
Hello, everyone. Welcome to another edition of the Metis Workjobs. We have been doing this every Thursday and it has been an amazing experience. We have been learning a lot about AI agent frameworks, how to integrate this with blockchain, how to use it on Metis. And today we're going to be talking about a very hot topic, MCPs with NIVI, right?
This is a topic that has been taking the AI space by storm.
So we want to really clarify what it is, how you can use it, and how does all of this connect with, you know, WebTree and blockchain and the things that we are working at at Metis.
and the things that we are working at at MEDIS.
So I would 100% recommend you to please share this with others.
If you are on Twitter, give it a like or a retweet.
And if you are on YouTube, we always appreciate your questions,
your comments, your likes, and that you subscribe to our channel, right?
We keep building this community around education
and we are pushing hard to create the best kind of content
that you can find so that you have all of the tools available to build amazing decentralized
applications on medis remember that we are on the year of regenesis in which medis is fully
embracing artificial intelligence and this comes with a series of steps and incubations and the launch of a new chain
called Hyperion.
So, in order to learn more about this, you can visit our website at medis.io.
But today, again, we're going to be focusing a lot on added, authentic framework, something
that we have been exploring thoroughly in several workshops.
And we're going to continue working on this as we build our community of amazing AI builders. How are you doing today, Nidhi?
I'm doing good and super excited to share what we have about the MCPs using Elite AI Agent and
all. So yeah, super charged and super excited to share what we have.
Yeah, absolutely. I'm very excited as well.
So maybe we can start with a small presentation about the content that we will be exploring
today while we give a little bit more time to the people that are just tuning into our
streaming.
Okay?
Yep, sure.
So let me share my screen quickly for this presentation.
Olga, is it visible?
Yes, I can see it.
So today we're going to be talking about unlocking the power of MCP for AI agents,
specifically integrating MCP with Alit and Last AI within the MEDIS ecosystem.
Super exciting topic.
I hope that everyone enjoys.
And if you're just joining right now,
I will kindly ask you again to give it a like, a retweet, comments,
and ask as many questions as you might have.
Go ahead, Nidhi.
Yep, let's get started.
So hello, everyone. Welcome to the live stream of the workshop series that we'm super happy to have you all here today.
We are going to dive deep into something
that I think a lot of developers
are going to find really useful,
something called like MCP.
So let's get started.
If you are hearing, I mean, MCP uh and thinking um okay is this another buzzword in
the ai or in the blockchain uh world uh i think that's okay all of the questions but trust me
this is one of the uh you know those tools that actually uh helps you to do more with
really helps you to do more with less friction.
So especially when you are working with AI agents,
it's going to really help for you.
So let's get started.
What is MCP?
What is actually the model context protocol?
So if you have ever tried to get an ai agent like cloud or gpt
and uh if you try to interact with files uh apis or a blockchain data
sorry so uh if you have ever tried to get the blockchain data APIs or read the files, so you know, it actually, how it usually involves a bunch of, you know, code and workarounds that, okay, there's like a lot of extra things that you have to do when you are especially working with ai so that's where uh the mcp comes in and uh mcp simplifies uh that whole process
like it is a standard way for you know ai to say that um say like um hey i need something for uh
and your system to respond like with a cool i mean okay with like okay cool here it is so instead of
um writing a new integration every time uh you can expose a simple functions using mcp
sdk and it is written for like different uh there is like different languages that's uh
there is like different languages that MCP actually supports,
but we're gonna check the MCP Python,
SDK for the Python.
So we're gonna dive deep into the like,
how we install MCP, how we create the project for the MCP.
Because if you are working with elite AI agent,
you need some different installations as well
for to integrate MCP in your own AI agent, you need some different installations as well for, for to
integrate MCP in your own AI agent.
So, so how does it actually work is instead of, you know, writing integration every time
you can simply just write a function in the MCP file in the server file like uh read this file get uh this latest price from this data
and and you just connect it and connection with the ai agent is also very it's also like very simple
uh it's just like you just define the um server config file and in your AI agent and you are all good. So it works with the stuff like your file system, your external APIs, your databases,
the blockchain data.
So we're going to see in a live demo coding that how it connects with the GitHub API,
that how it retries the data from the GitHub that how many folks
in this particular repository have list out all the repository that it this particular organization
have. So that's that's how I understand correctly Nidhi sorry to interrupt you there it's how I... So if I understand correctly, Nidhi,
sorry to interrupt you there,
it's just I want to make sure this is extremely clear
for the people in our audience.
Like, I have an agentic framework.
There are several of them.
And these agent frameworks,
well, there are several features
that can make them more useful than others.
But the key one, I would say,
is kind of the tooling, right? The integrations
with tools that you get into it. And with MCPs, if I understand correctly, I have now a standard
procedure on how to call a bunch of other APIs and other functions so that I can expand the tooling
of my IGN-TIC framework and do a bunch of things. i correct here yes yes for i mean definitely so
yeah it definitely helps like uh particularly in the ai agents cause like it has like a lot of
different use cases as well it might have the different uh use cases but what we're going to
talk about today is how we can integrate it with the ai agent So there are a bunch of things that your AI agent could not, I mean, just your simple
AI agent might not be able to perform all the tasks, like, I mean, to read this file
from this, I mean, from GitHub, from your system, like, so that's where the MCP comes
And that's why the MCP is important.
Like, you stop writing no repetitive code,
like no more manual code.
And it's efficient in integration
that it simplifies your integration flow.
Like rather than figuring out how your AI
can talk to every tool,
you just give it a consistent way to ask.
Like it doesn't care, like how things work under the
hood so it just calls the function and gets the result and you stay in the control uh of uh what
the ai can do and your you know workflow becomes more uh modular uh in terms of you know when you
write the code and it uh it becomes more flexible being a developer, getting this kind of tools
is like super, you know, pleasure to have.
And this is how a thing is structured in MCP agents.
So at a high level, if MCP gives you like three components,
like standard, standardized interface,
plugin interaction, and function call,
so all the tools follow the same format,
which is honestly refreshing.
So whether it's reading files, fetching data,
or something you build for yourself,
it works the same way from the
ai's perspective as well so that's why how they do have the standard uh interface and
if we talk about the plugin and integration uh mcp uh works with clients like typing mine which can help which can help route to route you know those requests
and response so to cloud or gpt for example say i want this thing and the client handles
the back end for so that's where this integration, you know, to like simplify the things and how it runs.
We're going to check and we're going to see how we're
going to see its core architecture.
So that's on that slide.
Maybe you will get more idea than how it's running
and how it is integrated in an effective way.
And if you have any questions?
Yeah, no, I think it's clear.
Yeah, I think the point is that to show people what they can actually use, right?
Because at the end of the day, we just want them to be more productive when using our AI agent framework.
And for that, MCPs are a key, right?
So I think it's great if we can talk
about the practical application.
Because like you mentioned here, it
simplifies creating integrations
and enhances the capabilities of AI tools, right?
And then next, we will need to connect this with Web3, right?
We are talking about Hyperion and, you know,
our new AI driven blockchain.
And of course, all of these tools are gonna be
of great interest and usage as we make progress, right?
Yes.
So if we talk about the practical applications
regard, I mean, in the Web3 world or out of the Web3 world,
stuff that you could try, yeah, like the viewers who
are watching, you can try this big.
Code review and file access, I see
that your AI can open and read the files from your project
folder imagine cloud checking your smart contract and you know supporting a misses uh missing access
modifier so that's what you can do uh using the mcp you can pull the live data that we
we're gonna check how we how we can pull the live data from the Web3. So I want to fetch like price feed user profile
or even social media data,
or you just register a function like get API data
and let your AI pull it in.
And you can also trigger actions like think in CICT,
your AI agent, your AI could trigger a build push to GitHub or send a Slack message when it finished checking something.
So what I love here is that, you know, these are the practical, not hypothetical ones.
So once the plumbing and all the things are set up. This flows are just a few calls away.
I mean, I mean, you know,
if you want to integrate it with the AI,
we'll see that how easy it is to, you know,
integrate with your AI agent using Elite.
So that's where you will actually see that
how amazing it is and how, you know,
interesting it is to play with it.
So when I started doing coding for all this thing together,
first I started doing to fetch the data from the Web3,
from the Metis blockchain.
Then I realized, okay, let me try to fetch the data from the github let me try to
fetch the data from my um my my system so when you imagine imagine just this thing that um you have
your own ai agent that you have created for yourself and you have set up like a set of instructions that do this do this uh line by line and it's uh it's
doing uh follow your instructions and making checking or doing data uh behalf of yourself
uh checking your file system so for me it's real interesting for if if you are finding this
interesting drop down the comments on the comment section and let us know that what do you think about this integrations uh this all three integration of a
powerful trial and how would you integrate it in your ai agent and what do you think about this
the trial mcp elite and laser got it perfect yeah i think this is going to be like, this is a great example of how, you know, AI is disrupting little pieces on the blockchain vertical from the data component to creating smart contracts to monitoring systems to, you know, improving the general experience for that that just like you mentioned here on this bullet point and what i think people should be paying attention to is the fact that all
of these components that might not seem very big at first uh we will see a sort of network effect
of ai disrupting this industry overall right so i'm pretty excited about learning more about mcps and how we can
use this within allied our agentic framework yep so um yeah definitely leave your comment and let
us know what do you think guys uh so if we talk about this the powerful trio mcp elite and less ai uh with the patties so uh we do have the comment from the
cobbin mcp elite unless it's the best trial since shag and jack fisher for sure okay yeah true thing
so yeah as we mentioned that in the earlier slide that it gets you the real-time blockchain
data access, it's due to AI assistance to your smart contract.
It can enhance definitely the user experience on your task.
Integration with Metis ecosystem project for sure, and automated market highlights and
analytics.
This is for, I i mean you can fetch the
live data and you know uh perform your things so if we talk about the core architecture of this uh
mcp so before we jump into the live demo let me show you that how the architecture works and
don't worry it's not too complex uh i'll put it simple so here is the layout host client
and server so host uh is uh this is where your ai model lives uh maybe jpt cloud or in in a in
your desktop app or in your id and client is uh it sits inside the host listening for tools request
the host listening for tools request from the AI.
And server is, this is actually where functions runs,
like reading files, calling APIs, and interacting with the
blockchain, like whatever you want to do with the MCP,
it's going to put in the servers.
And how this works, so let me give you the flow.
Let's say the AI says says i want to call raid file
this the client sends uh this to the server and the server runs the function and send the results
back and the client delivers to the ai so uh when i say ai it means your ai agent like when you are
running your ai agent you you just give it a prompt by
yourself that okay get me okay uh get me the details of the repo in this repository so it
gonna send to the server the server runs the function and sends all the results back and the
client uh yeah server sends the results back to the client and client will deliver it to the ai and ai will
show you that okay this have the this this this details of this um here are the details of the
repository so time to get fun get to the fun part so let's get started uh Scan this QR code to pull the GitHub repo,
and it will be super easy.
So try to run code with me,
and if you find any errors or anything,
so I'll be here to get your know, get your errors sort quickly.
So let me stop sharing this.
Okay, one thing before we move on,
because I think this is extremely important for people to know,
like, how, like, newbie can I be in order to learn to do this workshop?
I think that you can do this with very basic programming skills, right?
So if you're just getting started to code, this is something that you will be able to use. this with very basic programming skills, right?
So if you're just getting started to code,
this is something that you will be able to use.
Am I correct, Nidhi?
Nidhi Nidhi I think, yeah, if you have started doing code
and if you don't have a lot of programming experience,
you will still be fine to run this code.
You don't need a lot of coding um experience but of course like you
must understand the python language uh i mean the basic python language to understand the code and
if you have to change something if you want to change modify the code uh you you might be just
you know able to understand that thing or else uh all things will be good. Thanks for clarifying. Let's move forward
then. Let me share my screen. Okay, so is it visible? I think so. Everything, let me double check here, because since we have the logo here, everything looks good.
Like you're sharing right now the same kind of screen from the StreamYard that we are using, but all looks good.
If that's what you're trying to share.
Ah, okay, now we can see clearly the coding. Coding, okay, let me get the repository
first. Of course. So for the people that are watching us, please make sure to subscribe to
our YouTube channel and share this live stream with others so that we can all learn together.
and share this live stream with others so that we can all learn together.
2025 is the year of Regenesis and Medis and we're fully embracing AI. As part of that,
that's why we are doing these educational initiatives and as Nid is showing on the screen,
you can actually go to Medis Edu on GitHub and find a bunch of content and repositories that you can simple copy paste and learn a different variety of topics from deploying smart
contracts to AI integrations and today we have this wonderful MCP repository so
tell us more about it Medi. Yep so in repository, we do have the GitHub, the server files, basically the main file, the main AI agent file and the server config file.
We'll see how to set up everything because when you integrate MCP with your AI agent, it's like a bit in a different way.
So let's get started. We'll talk about MCP, Elite and UV here as well. So let me
explain what is UV. As mentioned here, UV is a lightweight, user-friendly and flexible Python
tool which is designed to manage your virtual environments and dependencies and projects in
a streamlined way. So when you want to use MCP specifically, the MCP SDK for the Python specifically, you have to create the UV and you have to install the UV and set up the UV project first and then we'll get the MCP files and the AI agent files.
So I already have installed UV in my system.
Let me quickly open.
So go to, I already have the UV in my system.
So what you can do, use this command to install
and I will, okay, I have forgotten to attach
the UV website link over here.
So what I'll do for the Windows, I will attach the UV website.
You can just simply go to the UV.
This is the basically the, and yeah.
That's the main page, right?
Yeah, this is the main page.
So for Mac US, just run this command and you will be fine. And for US, just run this command and you will be fine.
And for Windows, just run the command and you will be fine.
Yeah. Okay, we might want to add that to the documentation.
But for now, maybe we can just paste it here on the comments section.
Yes. Yeah.
So, and this is the MCP Python SDK that we have used.
And that's, if you're building the same thing,
you have to go through the same process.
So here is the installation process for the MCP.
If you are running for you from yours by your own self,
then you can follow this repository as well,
which from this repository, it will make a jump to the uv file so if you want to
read more you can go to this both uh repository and website and verify your installation whether
uh running this command or whether just running uv a test dash version to check the version of your
test dash version to check the version of your um you okay so let me make it a little big now it's
good so let's get a new repository and if you if you don't want to go through all the process but
i would recommend that uh to follow what is written in the ritme file because if you just
what is written in the ritme file because if you just uh fog the um repository this one i
please try to set up uv and mcp first by yourself and then fork it otherwise it's not going to run
anyway because you need uv setup and all so if you are if you want to run it for first time and
If you want to run it for the first time, by not getting many errors, try to run it this way, the way I'm doing.
So let's do demo MCP.
Okay, make your demo MCP.
So Nidhi, you might want to expand this view because right now it looks like very small
compared to the rest of this screen.
Yes, that's amazing.
Now I can see.
So for the people in our audience, guys,
you are still on time.
Just go for this repository
and start coding alongside us.
Yes, open the Visual studio code using the terminal by writing code space. Okay, let me make
it big as well. Okay, so we do have created repository. Okay, now we're gonna do UV in it.
We're gonna initialize the UV project in this file.
So what are you going to do?
Just pull up the terminal.
And do UV in it.
So it will get you the required files all here for your UV project.
And later, initialize the, I mean, create and activate the virtual environment your Do activate it later.
Okay.
So once we have the activations of the virtual environments, we're going to install the required
packages.
So we will need a request.
We are using this package to sign the request to the server
and to get the response from it.
And Web3 to fetch the data from the Metis blockchain.
Elite, of course, we're going to merge it with the AI agent.
So for the installation in the UV for project structure,
you need this command uv and pip install
and you can check i'm sorry you can check uh
you can check the uh what what the packages you have listed by maybe running this command
packages you have listed by maybe running this command uh uv list okay uv list so if you want
to make all that uh you have installed uh all the com um i mean we do have the elite packages
have the elite packages.
We do have the web three.
We must have the request.
Okay, let me clear the terminal.
And then add uv add mcp cli.
This is make sure that you add this thing
or else it won't get connected with your mcp.
So make sure to run this command for very sure
okay so after this you are supposed to run this command export export open api key go to the
let me show you go to the open ai
uh okay i think i hope this is the right one
okay i'll have to log in anyway uh you can use use the, we are using OpenAI as the model in the AI agent, so you will need API key.
Here I want to clarify something. What about having other options? Is that possible?
I haven't explored that one because I have been just using open API key.
I'm not sure on this thing that if Elite have like the different models supporting, but I'm...
We do have support for DeepSeek, for Lama.
There is a list of, you know, LLM models that you can check out on the documentation, actually.
So I guess, you know, we're comfortable with JITPT, and that's why we're, you know, using it on these tutorials.
But for people in our audience, you know, they have some other options if they don't feel great about this one, of course.
So, yeah, so if you, because OpenAPI, I do have the, I do have the purchase date.
And it's like not that expensive, like in like five, $10, you will get a lot of access
to it.
So I am super comfortable with chat GPT.
So I just use this one.
If you want to use a different model,
you will see in the AI agent file here,
you can just change the model and you can use the different one.
So make sure that you run this command by spacing.
If you're using openAI API key, it will start from SK.
So make sure you run this command. I'm not going to show you my API key, it will start from SK. So make sure you run this command.
I'm not gonna show you the my API key.
So I'm just gonna move to the next part.
So MCP, okay.
So now we're gonna create MCP server file.
We're gonna create two files, MCP server.
So, okay.
So now here, if you want to create a different,
you can merge into the same file as well,
but I prefer to create different files
for the different servers and for the different purposes.
So first let's create MCP server file
for web three fetching data.
So MCP server file,
and we already do have main.py file in which we are going to
put our ai agent code and we do have the mcp file
okay so let's uh copy and paste this one uh you can copy and paste from here as well as you can go
You can copy and paste from here as well as you can go to this file.
And in the file, we do have, I think, the latest one.
I will update the readme today.
So you will get the same code in the boot space.
And, okay, let's go to the main file and get the AI agent code and copy and just paste i'll show you i'll walk you through the code
and i'll show you what what you can change uh with space to change like the code how you can change
the code and once uh we do have all the things okay we do have created the AI agent code. We do have created the MCP server
file. And now it's the time to create the server configurations file. So for that, server config.json.
Let me copy this one.
This has the last one.
Okay.
Just paste it, save it.
And what are the next steps?
Okay, and you are done.
So I'll go to the different repository now as I
have set up the OpenAPI secret key in that file. So it is basically the same thing. Nothing else
added in that repository. So I want to show you. Okay, let me just open that.
show you okay let me just open that let me open that file cd my mcp project
okay let me make it a little bit big
little bit big.
Okay, so maybe this is a good time to do our recap
since we have over 300 people that joined at our session today.
So again, for the people in our audience,
if you're watching this, please help us spread out the word.
You can do it via like, retweet questions,
especially questions.
We really appreciate that.
And we are exploring MCPs.
We have three different scripts so far, Nidhi, right?
We have the main, which I guess is for the one that we are using for the agent. And then we have the MCP server code. What else?
We do have the main code, MCP server code. We do have the GitHub organization or server.
we do have the GitHub organization RG server.
So this code is basically fetched the data
from the GitHub API using the GitHub API.
So let me get started.
So how does it run?
So to run your AI agent,
you're gonna run the main file using Python 3 main.py.
And yeah, make sure you have the Python installed.
I forgot to mention this in the initial state,
but yeah, make sure you do have the Python.
We are all running this thing in the Python.
So, okay, make sure you run,
if you want to run AI agent, you run this command.
And if you want to run MCP server, I i mean any server just run python 3 mcp server dot pi so okay
so this is fetching the latest matis block number let us uh make sure that it's fetching the right number. Let's go to Andromeda and
okay so let us block is now 37 but when we run it was 34. Let me run it again.
Ignore this
red red thing. So okay so previous block number 37 and the recent is 38.
Let me read it.
It is not updated here.
I think it's, okay.
Yeah, it was not updated here.
It's just got, you know, from 37 to 40 direct.
So, that's...
And here we are on... Ah, okay. So so there is a minor lag right but the it seems that it's faster
doing it via the mc yes yes it's doing like you know faster way like let's 41 yeah it's showing
you the 42 it's just like four or five seconds ago so you know So you can imagine like how faster it's getting your things
from fetching like live data.
So this is, we have seen the output.
Let's see the server code.
So in the server code, we have installed,
sorry, import the packages.
We import fast MCP from using this to
fetch MCP.server.fastmcp.
Basically we are importing server to use these things,
initializing MCP server,
but the server name is Mattis block server.
And here we have connected Mattis mainnet API
to fetch the live data using,
if you want like to check the data of other blockchain,
you can just replace this one with that RPC URL.
And okay, so MCP tool is to,
it is used to send the request to the server
and to get the response from it.
So how you write it,
so you write it at mcp.tool and with the brackets
and define the function as usual
that how we define in the Python code.
And I have written this rhyme that just
explained that what this block of code do,
that it returns the latest block number from the MADIS
mainnet.
And here it is fetching the data using Web3.
Web3 means using this connection.
It will go to the eth block and fetch the block number.
So here in the that Web3 file, you see that it can fetch the gas price chain ID, get the
block number.
So like go to that and change the code here that instead of block number, fetch the latest price of the
metrics and let us know that how does it work. That would be great. And one question for you,
as there might be some people that, you know, are just getting started. What can I do with this
kind of thing? Let's imagine that you want to use this on production or for a decentralized application,
whatever, like something practical. What would you suggest?
I meant to run this application in the production level. So
you write the same code, like the same way, but depending on your code, depending on your requirements,
you must have to change the code a little bit for, of course.
So this tool, this is like for the very basic application.
So where I have used this guide to also have resource.
I mean, let's say if you are running or fetching data from the API, then you may use this MCPay.resource they have.
So this will, yeah, to decorate, to register a function as a resource. I mean, to provide the function can return here
that the text content and binary content,
basically to provide this data to your,
I mean, as the output, I guess.
So like they do have different format
to how this can, I mean, do things vice versa.
So for this, the basic one, we have used this tool one.
So if you go to that MCPSDK main repository,
they do have mentioned everything over there.
So guys feel free to go over there and check out
and let us know as well that what you are changing, what you are doing with this code.
And yeah, this is just a simpler one that I'm just doing minus one in the current block and returning as the previous block.
The previous one.
So and in the main file, you get the letters book just to print uh to call
this function and print it out and don't forget to mention this one mcp.run so that's where it
it will run this mcp server and get your things done so that's about the mP server thing. Let's jump to the main file.
So in the main file, import this all.
Maybe we don't need, let me comment it out.
We don't need the OS one, but yeah, import elite one and in AI agent.
Okay, I want to make sure, I want to show you one more thing here.
In the pre-member, what we write in the pre-member,
pre-member is like very, I mean, most, most,
most important thing when you are about creating the AI agent.
So let's say if I comment it out all, and let's say if I just,
you know, saying my AI agent that you are an assistant
that provides Mattis blockchain block numbers so it will give
me so now if I ask I do have the GitHub code here and it is also connected to the server I mean it
is connected with the MCP and I do have like two things your MCP server and GitHub thing. But when I have not mentioned here
that you are supposed to return or assist
and give the assistant about GitHub repository information,
blockchain related query.
So if I have commented down, it throws the error.
Sometimes it returns the things
if you have already done the things, if you have already
run the things, but it will give you the error in the end as well.
I mean, after even returning the things, it will not throw the error.
So if you are creating AI agent and if you are, you know, creating different server things,
make sure that you add here the important messages that
you are an assistant that provides Metis blockchain block number and a GitHub repository information
and blockchain related query.
So these things are very, very, very important thing.
If you are not mentioning here, even the last time we showed that when we were asking our AI agent to share us about the Python language,
it was not able to share as a Python language because we didn't say it that you are supposed
to assist the user about the Python language when we add here that.
Got it. So this provides like the basic set of instructions to the AI agent so that it knows which kind of support it has to provide and tasks that it needs to complete.
you have to pass the mcpconfig path and just put out the name of your config, sorry, JSON file,
so which we have the server config JSON.
We're going to check out this as well later after this one.
So in the main function, as usual, that I have put that it's running the print lines
and we saw this thing, the same thing on the previous one
as well.
If you want to check about this, the whole code,
feel free to go to the previous workshop,
which was about the basic of the AI agents.
So you will find out what's this all.
But let me explain in a very quick way.
So here you use the input
and whatever you will put here in the input it will store in the user input and it will pass to
the ai agent ai model and it will give you the response and it will bring the response to you basically just simple thing so
how do you connect your mcp with ai agent the answer is here you just have to set this up and
make sure you provide informations here and i think you are all good for that and in the server
config file what we mentioned so you first mentioned mcp servers and in this bracket you mentioned what
what all are the servers do you have so i think this is not for us for today so
but okay so for the blockchain so my server name is blockchain, command is Python
and argument is mcpserver.python. So whenever we want to run this server, we will just write
like python3 mcpserver.py5. So then it will run this. So that's just the basic commands
what you want to run. And for the GitHub org, we do have the same thing.
GitHub org.
So I'll make sure you have the both name here,
the file name and this name should be the same
or else you'll get the error.
Okay, so let's run the main file, main AI agent.
In file main AI agent.
So by the main dot PY.
Okay, so let's ask a lot of questions to this.
What is my blockchain?
No, what is Mereese L2?
No, what is my thing is I do.
Okay, so it's not about to give us the details about the Mereese L2 as we have not mentioned here that you are supposed to give us the information about the Mereese block.
block. Okay, let's run again and let's now ask about the blockchain block numbers. I wish it
runs. Don't worry, this is a live workshop. People understand that there are bugs that things can go
wrong. When you change even something little bit at the uh just before you will see the results in the
live workshop so where is the letters block number in my i see why it's not running okay so yeah, I have to set up the ENV file.
So here I'm going to stop sharing my screen for a while.
Absolutely.
And I will be back.
Absolutely.
No worries.
Take your time to set up your API keys.
So guys, obviously, we are not going to be showing API keys. We're not going to be exposing
any real API keys live, right? This is something that is personal. You should never do that. You
should never upload that to a public GitHub repository as well. Of course, this is a basic
security measure that we encourage everyone to take, right? So for the people in our audience, if you have any questions, please feel free to ask.
We are checking out the comments on the Medis developers account, on the main Medis account
on X, and also trying to check out everything on YouTube as well.
So if you haven't, by the way, please make sure to subscribe to our YouTube channel and don't
forget to give us a like to give us feedback because we want to keep improving this series
of workshops. We are doing this every single Thursday at 4 pm UTC, no matter what. Okay?
Nidhi, how are you doing? Still setting up that? Yeah, all good. Just give me a few seconds more.
Absolutely. We can give you some more seconds. Please stay focused there. For the people in our
audience there, we have Pangapanga saying Nidhi is the goal. I absolutely agree. Nidhi, you have
been doing an amazing job showing us how to go from zero to zero with
a Gentic framework, having to learn this from scratch, right? We are people that are on the
blockchain space and now we are having to learn about AI, everything from scratch. So it's been
quite, you know, interesting journey, I would say. And, you know, all of us are training
constantly to deliver the best kind of content to you guys.
Yeah.
How are we doing now, Nidhi?
All good, all good.
All right.
Is it visible, right?
Yes.
Okay, let's go.
Python 3.
Okay. What is the latest block number? Oh, nice.
Run, run, run.
There you go.
There we go. So here we have got 29130.
Yeah, you can check that out.
29130.
OK, we got it.
We got it.
So let's ask the different question.
What is the previous block number eight
mary's
meanwhile okay that's okay so yeah definitely
because it's yeah it's just it's updating live, right?
People are transacting on the blockchain.
Of course, each time this happens, we create blocks.
And this is updating in real time.
So you can see the effect there.
It's quite clear.
So let's ask this to list out all the reports.
So I'll show you one more thing that this line,
if I forget, please ask me the question about this line.
I have to mention that by how we manage when we,
how, I mean, I'll share, okay.
So let's first run this.
So can you list out all the repos from,
we actually don't need to give them the HTTPgithub.com. You can just write the
matis-edu as well. But here, here, here, here it goes.
All the reports that we have.
Okay, so what am I seeing here? This is beautiful. So this is all of the content that we have been
working so hard to put on our matis.do repository. Now people can check it out.
We do have the agent with MCP here. Mac, LTI agent demo. You see Mac, LTI agent demo, scaffold it's
too. Okay, now I see how it has been writing from the down. So you see all the repositories
been writing from the down.
So you see all the repositories it's fetching.
And now let's see the stars, stars of the repository.
So let's check the stars for this repo.
So it does have how many stars it does have three stars.
So can you get me the stars?
This repo has.
OK, I see. I see. This was the, in the prompt issue.
So can you get me the starts of this repo?
You need to provide with the information.
Yeah, I mean, in the first prompt, when we uh when we give the prompt the contribution the
report name was in the down uh on the next line that's why so a teaser is uh giving you the
stars three forks zero watches zero so
yeah
if we check this one. All good.
So this thing's working and let me get you to the code and explain what exactly the things are going on here. So in this GitHub server file,
we have important things, imported things,
and we have set up the server name, GitHub server.
We are using GitHub API.
You can get this from, I don't think it should run,
but yeah, you'll get that from the online,
I mean, from the GitHub official documentation as well.
In this function, we have,
this return the list of the public repository
names for the GitHub organization.
So from the URL you provide it get the github API from the github API
from your org name and how I mean basically how it's you know fetching the
URL and providing the response to you and providing the all the results in a
list format to you using this function.
And here we do have the GitHub repo stars seeing function.
How it's go to this URL, pass into the response and how we get the data from this response.
It's basically sent to the server.
Server will, from the AI agent, it will send to the server.
Using this function, it will send to the, using the GitHub API, it will fetch the data
and it will give the responded JSON version of the data.
And that's how we will get the things.
It's actually getting a lot of data from this things
like lot of details about the particular repository,
but here I'm just using this four things to show up.
And yeah, so basically the basic simple code of GitHub server.
But the thing is that when you don't mention about that, that use the GitHub repo info
tool directly with the correct owner and the repo name, I had to return this line.
So what was actually happening when I was calling the,
when I have not mentioned this thing.
And when I have not mentioned here,
this state, this function of call,
let me comment it down.
I hope you do have time.
Oh, we are running out of time time so let me be a very quick
and in the server file we do have this github orgs so basically what uh this do is that it
runs the uh github org server file but when when i don't uh add this here. And when I ask the AI agent to get me the list,
get me the number of folks or stars,
it was just getting me the list of the data.
But when I just run the stars things,
it was getting me, it's a lot of confusing guys for you.
I'm pretty sure, but if you guys are running
and if you see face any errors
uh related to this one for your different projects make sure that uh what you want to do uh how you
want your ai agent to uh you know check files to check this uh thing server this thing. You make sure that you add specific things here in the pre-mable.
It's most, most, most important thing.
Got it. Super clear, Nidhi. People here in our audience are cheering up for you. Great
stuff. Very helpful. And actually, these are winners of our bounty with Alit, right? So congrats to them as well.
Congratulations to them.
Yeah.
So since we are kind of on time, maybe we can do a sort of high level recap on what we covered today.
And, you know, just leave it like that so that people can check out the material.
The important thing is that people check out the material and give us comments.
And if they are building something, let us know because we are here to support you,
to build alongside you. So yeah, that would be super helpful.
Yep. So what we have in the plan is that we already did the basic AI agent workshop,
how you can build your own AI agent using the elite framework. Today we had the MCP one. We do have like two, three more
workshops on the AI agent using different tools. So you can go to the YouTube or the AIX,
but finding on YouTube would be more easier, I guess. So you can go to the YouTube and check
out the previous workshops to start this one as well and you can start with this one as well
so you can go to our matis dash edu github organization and check out all the repositories
we have as the tutorial over there feel free to be the open contributor if you have anything to
add in that repo if you have any feedback if you have
if you want to recommend us something feel free to reach out us and uh to open a pr request as well
we will be uh seeing it and we'll check and we'll definitely get back to you for sure so
back to you for sure so and to dive more deep into particularly mcp and elite integration
check out the mcpsdk python that official website that official gator repo you will check like you
see like a lot of things over there and important uh they have like yeah very good stuff written in a
explanation way so you can check that as well as the elite's official documentation
absolutely always before getting started with a new tool new framework whatever it is
uh make sure to check out first the documentation if you have some questions
you have us at 1 dm away uh thank you so much nidi for joining us today i think that next week
we're going to continue to explore uh the world of ncps nei agents because it's such a broad and
deep topic so many things that can be done and we really want to show more things that can be done
on chain on the side of dfi and things
that are actually very trending right now so that all of our builders can you know build amazingly
centralized applications on top of madis so thank you so much guys make sure to subscribe
leave your questions below we will be there to support you along the way all right yeah thanks
bye bye bye guys take care take care bye bye to support you along the way. All right? Thanks. Bye-bye. Bye, guys. Take care.
Take care.
