[Music]

Welcome back everyone. This is the

second episode of MCP and ADK series. In

the last episode, we learned about how

to connect an ADK agent to existing MCP

server and giving them new tools and

real world capabilities. For today, we

are leveling up even more. We're going

to build our own MCP server with ADK.

That means any MCP compliant client not

just ADK agents but agents from all the

other framework can plug into your

tools. So in this episode we will cover

two parts. The first part is the core

component you need to build an MCP

server with ADK. And then we will cover

three different use case including first

build a simple API MCP server and then

building a database MCP server using the

MCP toolbox and the last building custom

MCP server for advanced scenario like

generative media. We also have a

detailed written tutorial from the link

on the screen so that you can try it

yourself. All right, let's get started.

The first part of the video is the

essentials of building an MCP server. So

what do you need to build your own MCP

server with ADK? At a high level, it's

about wrapping your ADK tools inside an

MCP server. And here are important

pieces. The first piece is the MCP

library and this Python library gives

you the framework to build servers and

the ADK tools to expose. You will need

to start with an existing ADK tools such

as function tool that wraps the Python

function or any other ADK tool that

already built. And the next important

piece are two server handler. Your MCP

server will primarily need to implement

two key handler. The first handler is

these tools. It tells client what tools

are available and the next handler is

call tool. It runs a tool when a client

calls it. Another important piece is a

connection mechanism.

MCP servers need a way to communicate.

So for local development or simple

deployment, SDIO is common. For scalable

production environment, streamable HTTP

is often pref preferred so it can run

over the network. And last but not least

is async IO. Both ADK and the MCP

library use Python's async IO. So your

server code will be async first. The

main idea is your MCP server acts as a

wrapper and translator. It makes your

ADK tools discoverable and lets other

clients call them in a standardized way.

And now let's get into the second part

of today's video. We will explore three

different use case. The first use case

is building an API MCP server. We will

expose load web page as an API for this

use case. Let's work through this

example together. Say you want an agent

or any MCP client to fetch your web

page. Instead of calling ADK directly,

you expose this load web page tool

through an MCP server. And here's the

flow. First, you create a server script

like my ADK MCP server.py. Pi and inside

it you can use the MCP server to define

list tools and call tool and in the list

tools you can expose load web page so

client know this is exist and in call

tool run the ADK tool when a client

calls it. And now to test it, you can

spin up a second ADK agent as a client.

And that client connects to your custom

NCP server, discovers load web page, and

can fetch a website on demand. You can

also test it with ADK web. You can add

an init.py

file in the same directory.

Run adv from the parent directory.

Your ADA client will launch MCP server

as a subprocess, connect to it, and

fetch the page. Boom, you just turned an

ADK tool into a standalone MCP service.

Congratulations. Again, if you want to

follow hands-on written instruction, you

can also follow the link on the screen

or scan the QR code for more

step-by-step tutorial. And now let's

talk about the second use case. Building

a database MCP server. Imagine your

agent needs access to enterprise data.

But writing a custom database

integration from scratch is quite a lot.

That is where MCP toolbox for database

comes in. It is an open-source server

built specifically to expose database as

MCP tools. Instead of building

everything yourself, you can deploy the

MCP toolbox as a service and then

connect your ADK agent using MCP2 set

just like what we talked about in the

previous episode. And now your agent can

securely query and analyze data all

through this MCP abstraction layer. If

you want to try yourself, you can use

the link on the screen or scan this QR

code to learn more about this use case.

And last but not least, let's talk about

the third use case, custom NCP servers.

Sometimes you want to expose tools that

don't fit into the standard categories.

For example, generative media like

imagen for image generation, V for video

generation. Google even provides

open-source MC servers for those cases.

You can check the link on the screen

that provide more details. Another

example can be internal company APIs.

You will need to wrap your private REST

APIs and make them usable by any other

MCV client. Also for custom logic, for

example, you can add tools to do

mathematic calculation. You can define

the logic and make them a custom tool

for your MCP client. The process is the

same. Use MCP library. Define list tools

and call tool and pick your connection

method like stdio for local streamable

HTTP for remote. If you want to make it

easier, check out fast MCP. It is a

helpful library that simplify building

MCP server and making deployment easier.

By building custom MCP server, you can

really unleash your creativity. You're

no longer limited to consuming existing

tools. Instead, you're contributing

tools to the wider MCP ecosystem. All

right, let's wrap up for today. First,

building an MCP server with ADK means

wrapping your tools with the MCP library

and exposing them through list tools and

call tool. We also covered three use

cases. First, building an API MCP server

by exposing a simple API tool with load

web page. And secondly, building a

database MCP server with MCP toolbox.

And thirdly, building custom servers for

advanced scenario. By building your own

MCP server, you're not just empowering

your ADK agents, you're making your

tools available to the entire MCP

ecosystem. And that wrap up our two

episode series about ADK and MCP. You

now know how both connecting ADK agents

to MCP server and also building your own

MCP server. This is how we move from

isolated agents to a truly

interconnected AI ecosystem. If you want

to learn more details, make sure to

check the tutorial link in this video

description. Thank you so much for

watching and I will see you in next

series. Bye.

[Music]

Heat. Heat.

[Music]