Hey, how it's going guys? In this video,

we are going to see how we can build our

own cloud code assistant, right? You

would have been using cloud code. Cloud

is by Enthropic. So, it also helps you,

you know, in your coding workflow or

when you are developing something,

right? So, in entire coding workflow, uh

it helps you to write code faster, build

applications, so on and so forth, right?

also part of the same w coding uh you

know the terminology that we know. So

that is fine. You would have been using

cloud code but how can you build your

own cloud code right? Uh that's what

we're going to do in this video. So if

you look at here on my screen, we have

this highle architecture diagram that I

will explain you because it's very

important for you to understand that how

it's going to work first and then I will

show you the demo and all and code is

completely for you know so you can take

this code uh make this better improve it

and do whatever you want to do. It's

Apache 2.0 license so you can do

whatever you want to do with this. So if

you look at here this is the GitHub

repository and the repository name is

your cloud code. So we are building this

you know using lang graph and MCP

servers. So lang graph we are using for

the orchestration and MCP servers that

we are using you know to use some tools

via uh you know MCP servers like local

file access directory access read access

uh duck go for web search and so on and

so forth. So if you look at here on this

user interaction layer you know here in

the user it kind of uh uses a CLI. So

this is based on CLI how we use cloud

code. Right? So on the CLI it's very

colorful. I'm using reach CLI for that.

And then once you interact with it is an

application layer. In the application

layer this is the main.py file. So this

is your main.py file. It says the main

entry point for the cloud code

assistant. A minimalist AI coding

assistant using langraph and MCP model

context protocol. Right. Uh very simple.

So you can see the reachy thing is

happening over here you know the

initialize agent uh it's all

asynchronous right you try it out

because there are multiple thing that's

going to happen at the same time uh and

you can see it over here right uh the

reach UI that you are using so this is

the mainpy file which is the entry point

it initializes the core agent so there's

an agent dopy file the core agent of the

application if you look at this agentpy

file here right uh just go up it says

core agent implementation using lang

graph and mcp we are loading everything

so you know we are getting get local

tools get mcp tools you know we are lang

lang graph langchen etc we have a

console so we initializing the console

we have an agent state because you know

through lang lang graph we are managing

all the states right so if you look at

here this is the core agent and then we

we have a state graph that compiles that

graph because lang graph is compile

compile the graph Right when you work

with Langra. So over here if you look at

this it has an agent state which is your

dictionary. It says type dictionary. You

can see the agent state. It has update

queries response persist invokes invokes

like the everything that we can do

within a langraph workflow. We are doing

it over here. Now this lang graph

basically manages multiple states. So

you can look at the core components. We

are using anthropic APIs cloud sonet 4.5

as an LLM. We have an SQL DB that

persist all the all the database related

thing that we persist and then it has

some local built-in tools. So we have

some tools that are locally built and we

use that in MCP tools optional like you

can also use Doug Duggo. You can also

use GitHub and etc through MCP and I'll

show you that. If you look at it, it

says local tools. What are the local

tools we have? We have run piest. So you

can also do you can also run some you

know test cases on your files etc. We

have search files. It can find out all

the files. It has get file info. So it

can find out the information about

different files. You know it has read

file. It can read some information from

the files. Uh it has list file. It can

list down all the files. It has write

files. It can write files. So it

performs all of these locally. Okay. In

your directory and also has MCP servers.

You know in the MC MCP servers we have

duck go search like if you want to

perform web search you know through

duck.go or GitHub API through personal

access token or pat. Right. So this MCP

I'll show you that in a little bit. So

this is a highle architecture very

simple anybody can understand nothing

technical garbage over here. Right. So

I'll go back to the code. So if you look

at here this is we have an class agent

state. Uh we have a base model

initializing it basically. Right. If you

look at here it says minimalist AI

coding assistant using lang graph and

MCP. It says state graph with three

nodes user input model response and tool

use. That's what we having three nodes

that we having. So lang graph is based

on nodes and ages. Then you compile it

and then you know run that uh your agent

workflow. So if you look at it says

persistent state using SQLite

checkpointing. So we have here SQLite.

If you look at this checkpoints DB in

the left hand side I'm using winds of

other agentic IDE just like VS code. Now

if you look at this checkpoints DB

that's where we persist everything that

we are doing over here. Now here we have

our init function. Of course, this is

fine. You know, we are like initializing

everything and like we have a we have a

dictionary that basically says stored

number options from bullet points. I'll

show you that in a little bit. Here we

are initializing the LLM and you can see

we initializing our chat entropic model.

So net 4.5. We have some tools that we

are initializing. We have local tools.

We are getting the local tools and I'll

show you from where we are getting these

local tools in a little bit. Right. And

then loading the MCP tools, binding

tools. We have some display options. Uh,

you know, we can make this better

actually a little bit. Excuse me. Uh,

this can also come back once.

Yeah, this can all go back.

Exactly. Or something like this. So, we

have this cloud code assistant. We are

setting up the workflow. You can see set

up the state graph workflow. Uh,

conditional routing. You know, we have

async def initial that initializes it.

We have a SQLite check pointer over

here, right? Uh that looks at our uh

local database that we have. It it try

to load the MCP tools. If S gives an

error, it gives an error. Compile the

workflow with recursion limit. It goes

in loop. So it has to set a recursion

limit over here. This is what we are

doing. These are all the terminal

things. I'll just go quickly. Model

response. You can see there's a simple

prompt. It says you are a minimalist AI

coding assistant. Your capabilities,

critical rules, blah blah blah. And you

can improve this uh by your own and you

can see the tool use and everything. So

this is our agent.py file. Fantastic,

right? This agent.py. You can also add

more things if you want. Now this is the

agent.py file. I already shown you the

entry point. So this is our entry point

main.py. Here we are writing all our MCP

tools. Let's see that guys. Right? It

says MCP tools integration. It supports

desktop desktop duck.go GitHub sandbox.

Now here we have get MCP tools. It says

load and return MCP tools. It has

desktop commander which is your all file

operations right? You can perform CRUD

applications you know locally if you

want remove the delete part but you can

do the CRU part create read update. We

have duck go web search that can search.

We have GitHub for repository management

and then we have sandbox python to run

python code to do testing so on and so

forth and you can see it over here.

We're using MCP toolkit from langen mcp

and we're writing all the commands.

These are all npx command that you see.

Make sure you have node installed. You

just do uv sync you know that will sync

because we have used uv to set this up.

So not a problem. duck dug go mcp web

that you see how we are using it from

over from here. Uh github mcp you need a

github pad token that I already have in

myb file that I cannot show you right

now. This is our mcp tools. Pretty

simple. Right now local tools you know

you can see the local tools written in

basic python as a decorator that we are

using over here that you can see. uh

pretty much standard things that we have

right how to read how to list how to you

know access files how to write files and

so on and so forth right simple thing

nothing fancy it's just about how do we

orchestrate these because orchestration

is important right so this is what is

happening uh we also have a uh MCP uh

Dino MCP over here that you see sandbox

python execution it says based on pyic

run python you can also run that uh

through a docker container I already

have that here you know in my system. So

I also have given this file. Now how

does this all work? Like I've been

talking for a while. I've been showing

you this first time. I'm showing you the

code first. Otherwise I would have shown

the demo. But this is a bit technical

video guys. It's not for like people who

don't understand code right. You have to

understand MCP and all to go through

this video. It's not for like pretty

beginner. It's for intermediate people.

Now what I'm going to do is I try to run

this right. That's that's the agenda of

this video and see how it works. So I

already have you can see I'm running UV

run. Now what happens is the moment you

run it says what would you like to do?

List all files in current directory. So

available tools run the test. Let's say

I just do first. It says list all files

in current directory. You can see

selected. Now it list down all the files

that I have. See here it says content of

this. It has listed down everything.

Executing tools. it executing that tool

list files then you can see list all

this is how it comes up let me first

show you that so this is where we are

running it right eval uh u run mainpy

the moment you run it it shows cloud

code cloud code assistant not cloud

cloud cloud code assistant uh you can

also make this a bit fancy it gone that

side it says minimalist AI coding

assistant powered by langraph plus mcp

type exit or quit to terminate So but we

are not going to do that. Now if you

scroll down a little bit it says loading

tools. So it's loaded the six local

tools agent in a slice successfully. It

says what would you like to do? I'm

saying okay list all files. So available

tools run the test read the readme file

search for code in the project and I

have given one as a choice and the

moment I have given this choice you can

see it has given me all the content you

know the tool result of this list file.

So this is one thing right. The the next

thing we can also do is you know you can

ask of of course you can do a lot of

other things actually here. Okay. So let

me just clear this terminal for you and

I'm going to again run this and you can

keep on going uh here you know so

available tools run the test read the

readme file and whatever. So let's say I

read the readmi files read the readme.md

file and you can see it's so beautifully

it has fetched the readme file for me.

So I'll just go and show it over here.

So this is the readmi file that we have

100% correct. Uh it's it's working

fantastically guys. Right now if you

look at here

I'll just go up. By the way this is my

entire yeah you can see the tool result

read file. This is my readme file. Let

me show you the readme.m here it says

cloud code assistant a minimalist AI

coding assistant built with langraph and

mcp. Now why I'm showing you this right?

If you really want to master this

technology, you cannot just be using

chat, GPT, Gemini, cloud, cursor and all

and say okay I'm an AI expert. You are

not an AI expert. You are just a prompt

engineer, right? You know how to build

this tools, how to build something like

cloud code, how to build something like

cursor. Take VS code which is open

source. Take VS code, take some other

models, build aentic workflows, aentic

IDEs, see how compiler works, how do you

make compiler, how the operating system

works. You have to go that deep guys.

Trust me then you becomes then you

become a master of this technology

otherwise you are just a prompt engineer

or you are just a context engineer right

so it should not happen like that right

you have to go deep dive and then learn

these things how I've combined lang

graph with MCPS and build this CLI tool

just like cloud code of course this is

not like exactly like cloud code because

they have billions of dollars of funding

they have a set of team they can do

whatever they want to do I'm a solo guy

right so I'm just trying to see how we

can start because if you can start you

can really you passionate about it and

you can go and exactly build it. So this

is what it is right uh over here. Now

everything is given. Let's say you want

to visualize the workflow. I also have a

mermaid diagram generator that can

generate the diagrams and you can go go

and learn about it. So this is what it

is. You go and get this code try to

improve it over here. This is already

the architecture I shown you. You can

also bring up MCP tools. You can also

have local tools. Call whatever you want

to want to do. It's it's up to you.

Right? This is where you'll find out the

GitHub repository. You know the code is

completely for you. Go and use it. If

you want to contribute to this project,

you can also contribute it. Okay, feel

free to contribute. This is what I

wanted to show you guys in this video. A

quick video of how Langraph can be used

as an orchestration framework that can

use multiple tools either through Python

functions or you are using through MCP

tools. combine it together build cloud

code a CLI based code assistant for you

guys right that's all for this video let

me know if you have any question

thoughts or feedback in the comment box

if you have any other feedbacks you can

also reach out to me through my social

media channels find those information on

channel banner and channel about us I'll

also give you an agentic toolkit link in

the video description if you want to

find out a learning road map or if you

want to you know get some toolkits and

frameworks around AI agent and agentic

AI that's that link is available in this

video description. If you want to buy

readym made projects let's say some SAS

tools and some AI tools and want to sell

it in US or anywhere else in a higher

price. I also have a bundle of that

which is available around $1,000 that

you can take it and you can sell it for

$10,000 if you have the connections and

leads to sell it. That link is also

available in the video description.

Right. If you want to get involved in

any kind of AI training, we're also

doing that in physically, you know,

offline in class in Hyderabad and also

online. Comment that and we'll share you

this information. That's all for this

video. Thank you so much for watching.

See you in the next one.