A Brief Overview Before Anything Else: Artificial Intelligence and MCP

When the training of an artificial intelligence model is completed and its version is released, it relies on the information it was trained on when you use it.
But what happens if you ask it about something new that it was never trained on?
Let's assume the latest version was released in early 2026, and we are now at the beginning of 2027. An entire year has passed, during which many changes and developments have occurred. So what does artificial intelligence do when its information becomes outdated?
It may resort to guessing or searching the web, and it may be right or wrong. This is where MCP comes into play first.
Secondly, artificial intelligence is like a book. It needs a first edition, a second edition, a third edition, and so on.
AI models must be continuously updated, and this is one of the biggest challenges companies will face because updating a model is not limited to improving one aspect. It requires comprehensive development across various aspects.

First: What Is MCP?

It is like your own personal bag containing information, skills, and tools tailored to your specific needs.
For example, let's assume you are a computer engineer who is highly skilled at your job, but you lack the necessary skills to use certain applications.
This is where MCP comes in. It fills the gaps, much like someone taking online courses to improve their skills.

Second: Uses of MCP

As I mentioned, MCP can serve as a complement to existing tools and skills, and you can also integrate an artificial intelligence model within it.
You can also use it as a bridge between an AI model and various applications, such as Photoshop, 3ds Max, and others.

Third: How to Build an MCP

The process is simple. Start by gathering the resources you want.
Example: I want to create an MCP server or component specifically designed for drawing and graphic design.
In this case, I need to collect my own skills, books, references, indexes, and everything related to design. I can also include an internal drawing engine and a complete set of tools.
After that, I place all the resources in a folder, choose one of the most powerful AI models specialized in programming, and give it the following instruction:
Examine the specified folders, collect their contents, and organize them into a single MCP component.
Once the process is complete, I begin testing the component, improving it, identifying errors, and fixing them.

Fourth: Important Tips

1. Define the component's specialization: Understand what type of component you want to build, and avoid mixing unrelated tools.
2. Dedicate each component to its own field: Do not try to create one component that covers everything. Create one for design, another for programming, a third for writing articles, and so on.
3. Avoid excessive expansion: Be careful about adding hundreds of tools to your component system without a genuine need.
4. Make exceptions for complex projects: If the component involves complex engineering and requires a large number of tools for development and integration, you can expand it according to the project's actual requirements.

My Experience Building a Custom MCP Server for Drawing and Graphic Design

I had been struggling with repeated errors when generating images using artificial intelligence models, especially when including Arabic text in the images. The text would sometimes appear in strange, overlapping fonts, with distorted letters and results that did not match my instructions.

Therefore, I decided to build my own custom component under the name:

ArtForge AI MCP

First: Component Contents

I gathered a collection of specialized skills, references, and tools for drawing and graphic design, and integrated them with an internal open-source drawing engine.

According to a direct inspection of the component, it currently contains:

• 7 core design skills, including posters, cards, covers, social media posts, and more.
• 30 instructional guides covering design, drawing, and element manipulation.
• 9 specialized books and references covering design and visual arts.
• 351 indexed artistic concepts extracted from the references.
• 74 color palettes ready for use.
• 21 design systems for organizing visual styles.
• 138 drawing operations and low-level tools supported by the drawing engine.

Open-source drawing engine:

@napi-rs/canvas

It relies on Skia and HarfBuzz technologies, which help render graphical elements and shapes while correctly handling Arabic text, writing direction, and character shaping.

Second: Building the Component Using an AI Coding Assistant

I collected the books, skills, references, and drawing engine into a dedicated folder on my computer.

Then I selected an AI coding assistant specialized in programming and instructed it to inspect the folders, analyze their contents, and build an MCP component that combines the tools, skills, references, and drawing engine into a single system.

After completing the initial build, I began testing the component, identifying errors, improving its functionality, and gradually fixing the problems.

Third: Creating the Tunnel

After building the component locally, I needed to connect it to ChatGPT so that it could be accessed directly through conversations.

The connection process involved the following steps:

1. Launching the MCP server and drawing engine on my computer and verifying that they were running locally on ports 4273 and 4173.
2. Configuring the tunnel-client utility and creating a dedicated configuration file for the component named artforge-ai.
3. Creating a tunnel through the OpenAI platform settings and obtaining its tunnel identifier and connection key, while keeping the key securely protected.
4. Configuring the tunnel to communicate with the local MCP server, then checking the connection using the command tunnel-client doctor.
5. Starting the tunnel using the command tunnel-client run and verifying that it could successfully reach the MCP server.
6. Opening the component settings in ChatGPT, connecting the component through the tunnel, and testing whether its tools were visible and callable.

Fourth: The Result

I now have my own custom MCP component for drawing and graphic design, combining skills, references, a drawing engine, and programming tools.

The purpose of this component is to reduce recurring errors, improve Arabic text rendering, and provide greater control over drawing and design operations instead of relying entirely on conventional AI image generation.

The main idea behind my experiment is that I did not need to build a new artificial intelligence model. Instead, I needed to build a specialized ecosystem of tools and skills that an existing AI model could use to perform design tasks.

This is one of the practical applications of the MCP concept I discussed earlier.

Drawbacks and Observations

The component is excellent and powerful, but the quality of the output depends on the AI model used. I tested the component with several models and noticed significant differences in the results.
No matter how many skills you add, detailed instructions remain essential, especially in design, such as specifying fonts, colors, dimensions, spacing, and required elements.
Giving the model complete freedom may produce a good design or serious visual errors, and a single mistake can ruin the entire project.
This also applies to programming components, not just design.