Docker MCP: Packaging & Deploying AI Tools with Model Context Protocol

Hello, my name is Kristiyan Velkov and I'm excited to talk with you all today about Docker MCP, Packaging Deploying AI tools. Now, if you're not yet familiar with MCP, or Model Context Protocol, don't worry, we'll cover it in just a moment. I'll break down every concept related to Docker and MCP in a simple and easy to understand way. So, by the end, you'll feel confident working with Docker and MCP tools.

But before we dive into the details, let me start with a quick introduction about myself. I'm Christian Velkoff, a frontend developer with 10 plus years of experience. My passion is frontend and Docker. Also, I'm a frontend advocate, a tech blogger and author, and a Docker captain, a title given by experts in the field. Additionally, I speak at many conferences about Docker and frontend, topics I'm truly passionate about, and write numerous blog posts on these subjects. Here are the platforms where you can find my blog posts and content.

Okay, enough introduction. Let's talk about something really important today. What is Docker? Docker is open-source technology that allows you to create applications, images, and containers that work consistently across different environments. Nowadays, Docker is not just a tool for packaging applications but also MCP tools and running AI models locally. I'll demonstrate this with an example later on. And what about MCP? MCP stands for Model Context Protocol, a way for AI tools to communicate and understand each other, akin to the concept of a USB port C.

So for that reason, I decided to create a book about core utility types in TypeScript, something which is only one page in documentation. I did it in 600 pages. I give you promo code about my books. So if you go to my website, krishanvelko.com and type promo 20, you will receive 20% discount.

Okay, enough introduction. Let's talk with something which is really important today. What is Docker? Docker is open source technology. Docker is a way, most people know that, you can create application, use Docker, create the image, create a container over this image and everywhere will work the same. But, nowadays, Docker is not just that. Docker is not only a tool for you to package applications but also packaging MCP tools and also running AI models locally. I'll show that in the example later on.

Okay, what is MCP? MCP stands for Model Context Protocol. Shortly, it's a way how AI tools communicate between each other and understand each other. You can think of the concept of a USB port C. If you have a USB type C port, for example, you can use it in several devices, several tools, and later on, all of them can be connected to something which is an MCP client. Here it's more details about MCP. As I said, it's an open-source protocol created by Anthropic company behind the quote and also it's a really nice way to think about how something which is open source can change the whole idea of the AI today.

As I said, it's open-source protocol created by Anthropic company behind the quote and also it's a really nice way to think about how something which is open-source can change the whole idea of the AI today.

Okay. We already said that it's like the concept of the USB type C port, but also it's something which we can connect to the past. For example, MCP is like the HTTP or HTTPS protocol. They stand for something, right? They stand for how we share information. And here is the MCP, but how we share information for AI.

Shortly, how the MCP works. Usually, we have a client, think of the client like something you can write prompts or interact with it. For example, we can use a cursor, we can use a quote, we can use a client and many, many different tools. We write something to the client and connecting the client between the server and client with MCP, we can easily communicate with it. Also, this way is very intuitive because you can connect several tools, not only one, but you can do a lot of tools and interact in one place in the client. I will give you examples on that later on.

So Docker introduced something which is very, very powerful. Docker MCP toolkit, it's a gateway that lets you set up, manage and turn MCP servers and connect them to AI agents. So now, think of Docker not just like you have images, like Node.js image, Python image, but also you have an MCP catalog. So MCP catalog, it's a way where you can find how MCP tools integrate in Docker and just in seconds you have MCP connection. I will show that. It's very easy.

Now it's time for our code demo. To be able to work with Docker, you should install it. The easiest part and the easiest way to install Docker is in going to Docker docs, click on Docker Desktop, install for Mac for example, click install and follow the steps. It's very easy. I already did that. So for that reason, I will not follow these steps. But if you don't already install Docker in your machine, if you use Docker Desktop, it's the best easy way.

Since we are in Docker, what do we have here? First of all, you will see that we have our models and we have MCP toolkit. If you don't see MCP toolkit, what you should do is basically go to the settings, go to beta features and enable Docker MCP toolkit. After enabling it, restart Docker and you'll see MCP toolkit and also catalog here. Okay, let's see what we have here. First of all, we have our models. You can install AI models in Docker directly. For example, you can download Wama or Gemma or any available AI model in Docker. But let's take a look more in MCP toolkit. Okay, MCP toolkit is basically a way how I connect MCP tool with my client. So let's go in catalog. In catalog, you can see that Docker already added more than 200 tools. So for example, if you want to interact with SWAC MCP tool, you can do it by clicking here and just up MCP server. That is so easy. Okay, let's search for something. For example, Docker Hub. I already added Docker Hub, but if you don't, what you should do, it's very easy.

I already added Docker Hub, but if you don't, what you should do, it's very easy. Just click Docker Hub, add your credentials for Docker, I'll do the steps with you. So click again, Docker Hub, and that it. It's already added.

But, but, but, but, wait. So we already added the Docker MCP server, but how to connect to the client? Docker did that very, very easy. So as you can see here, after we decide what catalog we want, they will be listed here. For some of them, we will need to provide the credentials. The beautiful part about Docker here is that the credentials are not in the code, are not in some files, but directly live in Docker. So they are secured.

Okay, so what I have already is three servers. I have a Docker, Docker Hub, and SWAC. For SWAC, of course, if I want to use it, I need to add my configurations, like secrets. You can follow the documentation and add them, but for now, I will not do that. Rather, then, I will try to add the Docker. So let's remove Docker for a while, just for demo purposes, to show you how we can do that.

Okay, now I don't have any MCP server yet. So I click on the MCP server. I type Docker. I just click the plus button. That is, the server is already added. Now let's see how I can connect this server to my client.

As I can say many times, the client is just a tool which you can interact with and communicate with the server. So in my case, I will use cursor. If you are already connected, you will see that red button, but if you are not, just click Connect. Now we are connected. The next step is to go to cursor.

So here we are. We are in cursor. Now what you can see is that this is my demo application which I created for you. This demo application, what it's doing, is basically prompts. If you type what is MCP, this will give you the output what is MCP or type what is Docker, will give you information. We'll see the examples how it's using in seconds, but now the most important part is how to connect MCP server to MCP client.

In our case, we are in code or cursor. In cursor, you can do tools MCP, and if you are not added yet, you will see that we have these tools. I already add them. But in case you didn't, what you should click is basically cursor settings, new MCP server. In MCP server, we have MCP JSON, and add your MCP servers.

As you remember, we added Docker, right? So if I go to Docker, my servers, I have Docker here. Perfect. But why I don't see it in the MCP? Because often you should restart the cursor or the client you have. Okay, I will do that right now. Perfect. I restart the cursor, and now you see that I have MCP Docker added in the configuration of MCP JSON. So perfect. Good. We added it.

Now, let's activate the tool. Okay, MCP Docker is already activated by us. If it's not, come here to tools and MCP, and just double check it's activated or not. And after you added this, the beautiful part is that we have a Docker here. See? And this command, I can use it to do a lot of things. For example, Docker help. See? Docker help gives me all of the commands which I can run with Docker. Beautiful.

Let's add another MCP server. Return back to Docker. Add MCP server. Okay. I want to time. Add MCP server. My MCP servers, and we see that we have time. So, using MCP server, if I copy this code and paste it directly in cursor, it will work also. So, go in here, new MCP server, add the new MCP server, fix the mistakes. Done. Basically, like that. Basically, like that. Perfect. Now, we added the time.

Okay. Let's take a look what the time can do. Of course, we should restart again the cursor. Yes. Okay. Perfect. Now, as you can see that in Docker, we have convert time. And basically, let's see what we can do. Perfect. See? So, convert time is basically server which is used to convert the time. Short, it repeats that. It's not doing a lot, but as you can see in Docker desktop and the catalog itself, we have many, many different servers, which I can install and do a lot of beautiful things like scrapping the information, using Grafana or using Playwright.

A lot of things. Also, the beautiful part is that you are not limited to these sets of tools and servers here. All you have to do is create your own MCP. And that is exactly what I have here.

Okay. I named this repository AI code submit MCP Docker and will be available in my GitHub account. You can go download it and start basically running the commands. You can find the repository here. So, Christian Velkov, AI code submit MCP Docker. And I added very beautiful documentation on each step which I follow. In that way, you will see that running the MCP is not so hard.

That being said, going back to the application. I will show you an example of what this application is doing. And later on, I will show you how to create MCP. Okay. I type info.mcpserver. What is MCP? And it is asking me for permissions. After receiving the permissions, you can see that it's typed the response. What exactly is MCP? I did this MCP server just to demonstrate how easy it is to create MCP. But also, if you forget what is MCP or who I am, you can type a comment and it will appear, for example, my name. Like this. Running the command. Perfect. We see output.

Nice. Christian Velkov, from the developer, experienced. Where you can find me, where you can connect me, where you can ask questions. So this is beautiful. And this is the result. I just created this simple MCP server in seconds with help with AI and also with knowledge how to configure the MCP servers. And to be honest, it's not so hard. Let's see what we have. Okay.

Okay. First of all, you have a server. Usually, when I start creating servers, I use model context protocol SDK open source project and you can just download it and create the servers with them easily. It's open source. So just go and try it.

After I created the server, I set the information about the server and also the prompts which you can ask. As you can see, I had several prompts and based on each prompt, you receive some response here. I will not go deep how I did exactly the setup and the prompting because you have the option to go to GitHub and look everything.

But the most important part here is that in each step, you have a guide. And also, I added a course for mcp-config for you. I added a docker guide and docker compose, docker files, and etc. So usually, when you create a server, you can run this server locally. It's not a problem to run it locally. You can connect this local server to Courser, for example, as I did here.

You can connect this local server to Courser, for example, as I did here. See? This is server local. But imagine I want to share this server to someone else. The problem with sharing any information about MCP often needs security because you should share also how you did the authentication, authorization, etc. Also, I have a package.json file, which specifies different dependencies, specific Node.js version, for example, or different npx tool. So here is the beauty of docker. Docker allowed me to create an image for this specific server, and later on, to paste it in some repository like my private Docker Hub repo or anywhere else and share it with the people without worrying that npm will not work, or the packages are different, or the security issues. Everything will work if you do it with Docker. So for that reason, what I usually do is create an MCP server and later on package it to be an image, deploy it to Docker Hub and use it.

Let's show you how I did that. Okay, perfect. Here I add all of the steps which you need to follow to deploy your MCP server. Okay, let's just build the application. Done. After the building the application, we'll see that we have a dist folder. Okay, next step is to test if this is working. See, we have a result. Of course, a new one. But the result is that it's added information about the prompts which I can use directly in the console. Okay, now we are ready to build the image. To be able to build the image, you should have a Docker file and also a Docker ignore file if you want. I highly suggest you to have it because then the image will be really small. Okay.

Let's build this image. Perfect. The image is built. Now, going back here. To be able to push images in your Docker Hub, what you should do is to log in first. I already did that, so for me it will be not needed, but for you, if you are not logged, you should do it. And also, you will see that logging succeed if you already succeed logging in Docker Hub.

Next step. Let's tag, for example, our Docker image. I'll type here .02. Okay. Okay. The tag is created and just push it. But before that, let's just copy the push. Okay. Pushing the images. Now, my Docker MCP locally, it's put in image and this image is pushed to my private Docker Hub. So, I go to Docker Hub. Here. See? I already have info MCP server. Perfect. And you can see that I have tag 2, which I added before seconds.

Many of you probably will ask, okay, I managed to create the image. I push it to Docker Hub and then what? How to use this image in the MCP? The easy question, it's really, it's very, very good because in the end, what is the purpose to use Docker if you can't use the image directly in the MCP? But here's the tricky way. What we have is that if I go back then to tools and MCP and take a look, what I have here is that this is my local MCP server, which is with argument server.js. But if you look this one, I directly use image which I pushed before seconds. So, for example, let's change these to the specific tag. Okay. Perfect. Nice. Done.

Now, what this allow me is that rather than to use the code locally, I can use directly my generated image like MCP server. So, here I have two commands. One is for local and another is for Docker. As you can see, I already typed some questions and receive some answers. Okay. For example, who is Christian Rilkov and this give me the answer.

Okay. Let's copy this one. Okay. Now, this is server running locally. So, if I put here local, we will see that it's executing exactly the output of server.js. Okay. That is working. Okay.

Let's try now with the Docker one. Okay. Now, what I should do is change the name to Docker server. Docker. And let's ask again who is Christian Rilkov. Now, the information is around directly for the image which is living in my Docker hub and I put this image before two seconds. Perfect. See, the output is here. This is proof that if you dockerize the whole MCP server and push it directly in Docker hub, in the end, the result will be a fully functional MCP server from Docker image. Beauty.

Okay. So, let's see what else we can do in our MCP. For example, let's ask what is Docker? Okay. Again, asking for permissions. And here is the answer. This answer is coming for our server. Of course, packaged server in Docker image. So, if I search here Docker. Yeah. See? What is Docker? Also, we can ask what is Docker MCP toolkit. Okay. Perfect.

And this, of course, will ask again permissions and will give me the whole result here from our MCP server. Beautiful.

Okay. To be able to see the images, one way is here, Docker images, and search for your image. Christian info MCP server. See? This is the tag which I created and pushed to my Docker hub. And another way is to use Docker desktop.

So, in Docker desktop, go into images, and you'll see that I have this image. So, info MCP server latest. And I have also with tag too. Perfect. Absolutely amazing. Now, I want to show you something very beautiful also about Docker, and that is the way you can ask AI to improve your Docker files.

Okay. Okay. Let's close that for test purposes. Help me to my Docker file. Now Gordon, AI, will ask me, okay, perfect, I will help you, but will you please give me permissions to access your directory. So, we'll be understanding the directory which I have a Docker file. Later on, we'll scan the whole application. We'll see that I have a Docker file, and will add me suggestions what I can improve.

See? Beautiful. Gordon is AI tool which I use a lot because it's trained to help people to work with Docker better, and you can scan your Docker files. You can type to give you suggestions what will be better in your specific application, and asking several questions related to Docker best practices, etc. So, highly advise you to go and check it.

So, we did our training. Again, I will mention that this GitHub repository is public in my account. If you go here, you'll see that I added the repository. You can go, check it, research a little bit what it's doing and how you can work at it. The beautiful part is that I not only added the server but also explanation what exactly each file is doing. That will be for our code demo.

Now, let's just go and share with you where you can find me, where you can connect with me. For example, if you scan this QR code, you will directly be redirected to my LinkedIn. Again, MCP, Docker, all of the staffs are really easy to understand. Also, if you are using already created MCP server like I did, it will be easy for you to understand the whole concepts behind MCP and Docker. I tried in this demo to show you the basics, but this is not limited just to give you responses which is hard-coded.