Introduction

0:00
all right builders let's break down this
0:01
MCP thing model context
0:06
protocol you've probably seen the buzz
0:08
maybe on Twitter maybe in your dev
0:09
circles everyone's calling it the next
0:12
big thing like literally the future of
0:14
how LLMs talk to well you know
0:17
everything your databases your apps your
0:19
code even your crazy prompt chains think
0:22
about it remember when USBC dropped
0:24
suddenly one cable to rule them all
0:26
charging your laptop your phone
0:28
headphones connecting displays total
0:31
game changer right that's the vibe MCP
0:33
is going for but for AI it wants to be
0:36
the universal connector okay so we're
0:38
going to cover three things first thing
0:39
is what is MCP and why should you
0:41
especially if you're coding even care
0:44
second is the origin story how did this
0:46
become a thing what's the pain point
0:48
solving and the last thing is real world
0:50
action we'll jump into Windsor and
0:52
cursor solid AI dev tools and actually
0:55
hook something up using an MCP sounds
0:57
good all right so let's dive
0:59
[Music]
1:02
in first is what actually is MCP so MCP
What Actually is MCP?
1:07
is equal to model context protocol a
1:10
super simple idea but with crazy
1:12
implications think of it as a
1:14
standardized plug-and-play system for
1:16
all AI models gpd4 Claude Miscrol
1:20
whatever whatever hot new model drops
1:21
next week and the reason that these are
1:23
so important is so that it connects with
1:25
different data sources different tools
1:27
APIs you name it but why now let's map
1:31
the evolution so it kind of went through
1:32
phases right so now let's talk about
1:34
phase one just LLMs this was the start
1:37
chat GPT Claude etc explode onto the
1:40
scene insanely cool tech but let's be
1:42
real at their core they're just
1:44
predicting the next word based on
1:46
massive training data on their own
1:48
they're kind of dumb they hallucinate
1:49
and do some stupid now let's move
1:51
on to phase two llm plus context this is
1:55
the billion dollar step things got way
1:57
more interesting here companies realize
1:59
LLM plus external stuff is equal to
2:02
value for example Plexity LLM plus web
2:04
search boom let's talk about cursor it's
2:06
an LLM plus your actual code base
2:09
whether it's reading or writing and this
2:11
has insane productivity boosts let's
2:13
look at your Gemini tools which is an
2:15
LLM plus your Google Workspace which is
2:16
your emails your talks or your sheets
2:18
and suddenly it's useful for your daily
2:20
tasks these aren't just chat bots
2:22
anymore they're integrated tools and
2:24
that's why they hit those crazy
2:26
valuations they solve the context
2:28
problem and that means no more manually
2:30
spoon feeding data from 10 different
2:32
tabs and we all know AI is only as good
2:34
as the context that you give it now
2:36
let's move on to the third phase which
2:37
is your MCP fixing the integration mess
2:40
so right now it seems like LLM plus
2:42
tools equals problem solved right almost
2:45
what's the issue scaling this
2:47
integration is a nightmare it's a
2:49
classic N into M problem you have N AI
2:52
apps trying to talk to M different tools
2:54
databases APIs and every connection is
2:57
custom connect to Postgress custom code
3:00
specific authentication connect to Slack
3:02
different API different framework
3:04
different authentication so even if you
3:06
want to connect your local file system
3:08
you have to take another approach it
3:10
becomes a combinatorial explosion for
3:12
developers and that is absolute chaos to
3:14
maintain sure we had APIs web hooks
3:16
random frameworks but it was like having
3:18
a drawer full of different charger
3:20
cables for every single device
3:22
fragmented mess so MCP steps in and says
3:26
hold up let's just make one standard
3:28
protocol like USBC which is one way for
3:31
any AI to get any context that it needs
3:34
using a defined set of rules okay so
3:36
let's break that down let's look at the
3:38
model any AI model GPT4 claw 3 Gemini
3:42
future GPT5 text image video models it
3:45
doesn't matter now let's look at the
3:47
context the stuff the model needs to be
3:50
useful code from your repo data from
3:52
your database or info from Slack files
3:55
from your disk function calls basically
3:57
anything that makes the AI smarter for
4:00
your specific task a protocol is the
4:02
standardized rule book the common
4:04
language how the model asks and receives
4:07
information how the model asks and
4:09
receives context it's not a tool itself
4:11
it's the standard everyone agrees to use
4:14
think HTTP for the web with MCP your AI
4:17
can discover and use data sources local
4:20
or remote without you building a million
4:22
fragile custom integrations which is
4:24
huge now we're going to talk about the
MCP Architecture
4:26
MCP architecture okay you get the what
4:29
and the why now how does it actually
4:31
work under the hood super high level
4:34
right stay with me mcp client this is
4:36
your AI application or agent think of it
4:39
as the brain the thing that needs data
4:41
or needs to do something it's running
4:43
inside cloud desktop cursor windsurf or
4:46
your custom AI app it initiates the
4:48
requests now we move on to the MCP
4:50
server this is a lightweight helper
4:53
program and this part knows how to talk
4:55
to a specific thing using the MCP
4:57
standard got a local database run an MCP
5:00
server that speaks SQL to that database
5:03
want Slack integration run an MCP server
5:05
that talks to the Slack API if you need
5:08
access to local files an MCP server can
5:10
handle that next we're talking about the
5:12
data sources and the services this is
5:14
the actual stuff that the server
5:15
connects to and these are of two types
5:17
local and remote so let's talk about
5:20
local first this includes stuff like
5:22
your files databases on your machine
5:24
personal APIs that are running locally
5:26
and when we're looking at the remote
5:27
aspect of things this can be anything on
5:29
the internet whether it's Slack Gmail
5:31
Stripe Notion GitHub APIs cloud
5:34
databases so they all communicate using
5:36
the MCP protocol the client finds a
5:39
server that advertises let's say GitHub
5:41
capabilities the client then tells the
5:43
server yo using MCP rules create a new
5:46
repo called awesome project and then the
5:48
server goes I got you fam and interacts
5:50
with the GitHub API and reports
MCP in Windsurf
5:55
back all right so it's enough time for
5:58
theory now we're going to jump into it
5:59
so let's start by opening up Windsor and
6:02
then setting up an empty project in a
6:04
new folder
6:06
all right so once you're in the next
6:08
step is to go to Windsoroft settings
6:09
which is down right here and click on
6:12
the advanced settings option so in here
6:15
you'll see cascade and under that you'll
6:18
see MCP so go there and see if it has
6:21
written ad server okay so once you find
6:23
the ad server click on that and now
6:26
you'll see a bunch of templates so for
6:28
the scope of this video let's add GitHub
6:30
mainly for two reasons all right the
6:32
first one is we're seeing that GitHub is
6:33
becoming a nightmare for nontechnical
6:35
folks so secondly it's super super
6:37
useful for VIP coders who are not
6:38
technical and still want version control
6:40
to host using Versel Netify etc so now
6:44
the setup is relatively simple okay so
6:46
first what you need to do is create a
6:48
GitHub account so for that we're going
6:50
to go to
6:51
[Link] click on the sign up
6:55
button okay this process is super easy
6:57
takes about a minute after you sign up
7:00
you'll be redirected to the homepage but
7:02
as you know we need to give access to
7:04
this account to Windsurf so there must
7:06
be some sort of linkage going on so for
7:09
that linkage there's something known as
7:10
personal access tokens you need to
7:13
scroll all the way down and find
7:15
developer settings once you click on
7:17
that and then click on personal access
7:19
tokens now you're going to find these
7:22
fine grade tokens the reason we chose
7:24
that is it lets us control how much
7:26
access MCP has when we add it in
7:28
windsurf so all of that is under your
7:30
control and in order to get control of
7:32
that you need to do this so now let's
7:35
click on fine grained tokens and
7:38
generate a new token we're going to name
7:40
it MCP surf and set the expiration date
7:42
to this you can set it to whatever you
7:44
want we're going to put it 1 week from
7:46
now and scroll down so I'm going to
7:48
select all the repositories and then we
7:50
need to select the specific permissions
7:52
so this is actually very detailed and
7:54
you can really spend time thinking okay
7:56
what do I want to have access to but to
7:59
keep it simple just follow what we're
8:00
doing okay administration read and write
8:04
commit statuses read and write content
8:07
read and write and issues read and write
8:11
then after you finish that scroll all
8:13
the way down and click on generate token
8:16
again you can set any permissions you
8:18
want now just click on add server so
8:21
once you click Click on add server it's
8:23
going to ask you for personal access
8:24
token remember this is what we made
8:26
while making the new account so take
8:28
that same access token and put that in
8:30
over here this will take a few seconds
8:32
to connect that's going to take a couple
8:34
seconds to connect and once it's
8:35
connected you can create update files
8:38
search repositories create repositories
8:40
blah blah blah all of that stuff it's
8:41
pretty impressive right so now let's
8:43
test it out by creating a new GitHub
8:45
repo
8:48
okay so let's simply ask to make a new
8:50
repo named builders central and see our
8:54
account is new and doesn't have any
8:56
repos at all
8:59
okay okay i got one here so let's see if
9:02
this
9:03
works similarly we have Postgress we
9:06
have Stripe Slack and many more
9:08
templates over here and not just here
9:10
you can add hundreds of templates online
9:11
on sites like Composio which is like an
9:13
app store for MCP client etc how MCP
9:17
changes the game for VIP coding i mean
9:19
look at this we just set up a GitHub
9:21
integration without writing a single
9:23
line of integration code we just talked
9:25
to the AI in plain English that's the
9:27
dream of VIP coding right tell the AI
9:29
what you want and it handles all of the
9:30
grunt work the problem was that VIP
9:33
coding breaks down the moment you need a
9:35
new tool or data source that AI doesn't
9:38
know about you'd obviously hit a wall
9:40
needing real coding or some hacky
9:41
workaround mcp makes VIP coding way more
9:45
powerful and realistic you need to
9:46
orchestrate a workflow across GitHub
9:48
Slack and maybe Stripe connect the
9:51
respective MCP servers you just need to
9:53
focus on the what not the how of the
9:55
connections this isn't just for VIP
9:56
coders though even for experienced
9:58
developers this standard potentially
10:00
simplifies managing tons of integrations
10:02
making systems less brittle and easier
10:04
to scale but for those of us leveraging
10:07
AI to build faster maybe without deep
10:09
traditional coding skills MCP is
10:11
potentially insane it unlocks the next
10:13
level of AI orchestration the more
10:15
people build and adopt MCP the faster we
10:18
get the universal standard if you're
10:20
like us and excited about AI doing more
10:22
than just generating text or code and
10:24
actually orchestrating tasks across your
10:26
entire digital life MCP is something to
10:29
watch closely let me know in the
10:30
comments are you hyped for MCP all right
10:33
so that's the download on MCP from us
10:35
here at Builder Central and until next
10:37
time keep building