Designed by the author using Figma and AI

Stop Drawing, Start Prompting: Architecture Diagrams the Easy Way

Because ain’t nobody got time for Visio

5 min readSep 22, 2025

Here’s a simple truth:

if you’re still drawing Azure architecture diagrams by hand in 2025, you’re doing it wrong.

I used to waste hours in Visio and other tools.
Dragging icons. Aligning arrows…
Googling “Azure Load Balancer PNG transparent background”…

The End Result?

A diagram that was already outdated the second I talked to another architect or a customer.

Here’s the thing… we live in an AI-first world now.
That’s right I built azure-diagram-mcp, so I can just say:

“Create a diagram of an AKS cluster with 2 services behind an App Gateway, connected to SQL DB”

And boom 💥 No Visio. No stencil packs. No wasted time.

It’s a MCP (Model Context Protocol) server that plugs into Visual Studio Code, the best IDE out there, using GitHub Copilot (it’s free) and turns your natural language prompts into good looking Azure diagrams.

Yes, it’s time to make your life easier.
Let’s dive in 👇

Why do we need Diagrams at all? Like seriously?

Diagrams are the universal handshake in tech 🤝.
When words fail, you draw boxes and arrows.
But the old way of making them is broken:

Too slow → by the time you finish, things have changed.
Too manual → drag, drop, resize, align, repeat.
Too siloed → stuck on someone’s desktop, never versioned.

Meanwhile, the way we work has shifted:

Everything is as-code (Terraform, Bicep, GitHub Actions).
Teams expect speed and automation.
AI has become the natural interface for describing intent.

So why are we still diagramming like it’s 2005?

Time for an upgrade:

You tell the story. AI draws the picture.

How I approach It

Here’s my 5-step playbook:

1. 🎤 Talk, don’t draw

Tell AI what you want:

“Create a diagram of a web app on Azure: App Service + App Gateway + Key Vault + SQL Database.”

The MCP server translates that into Python Diagrams code and generates the PNG for you.

2. 📂Keep the diagrams in the repo

Store the diagram code + PNG in GitHub.
Now it’s versioned, reviewable, and never lost in SharePoint hell.

3. 🔄Iterate in seconds

Want to swap SQL for CosmosDB? Just say:

“Update the database in the diagram to CosmosDB.”

…and it’s done.

4. 🤖Automate in CI/CD

Pair MCP with GitHub Actions: every time the diagram code changes, the diagram pictures update. Your diagrams become living documentation.

5. ✨Use AI like a design buddy

Think of AI as your diagram intern.
You give directions, it does the work. You just refine.

How I use it 🎯

Here’s how the stack fits together:

Azure Diagram MCP Server

The secret sauce. The Mozzarella to your pizza dough.
An MCP server that plugs into your AI assistant.
Prompts → Python code → Diagram image.

Diagrams Python library

The engine that renders Azure icons, clusters, and arrows.

GitHub Copilot in Visual Studio Code

Use the AI right inside your IDE and for your repo to call the mcp actions.

My favorite workflow:

Describe architectue in natural language.
Let GitHub Copilot generate the initial diagram.
Tweak with prompts.
Commit to repo.
Done.

I literally did a workshop where we created an architecture live using MCP prompts. Everyone was blown away. What used to take hours in Visio took 5 minutes with AI.

Here is the prompt and first iteration:

“Generate an Azure architecture using the AKS Landing Zone to create an enterprise level AKS architecture diagram.”

"""
Diagram generation code for aks_landing_zone_architecture
Generated by Azure Diagram MCP Server
"""

from diagrams import Diagram, Cluster, Edge
from diagrams.azure.compute import AKS, ContainerRegistries
from diagrams.azure.network import VirtualNetworks, Subnets, ApplicationGateway, LoadBalancers, Firewall, PrivateEndpoint
from diagrams.azure.network import VirtualWans
from diagrams.azure.security import KeyVaults
from diagrams.azure.monitor import Logs, Monitor
from diagrams.azure.identity import ActiveDirectory
from diagrams.azure.storage import StorageAccounts
from diagrams.azure.database import SQLManagedInstances, CosmosDb
from diagrams.azure.devops import Devops, Pipelines

with Diagram("AKS Landing Zone - Enterprise Architecture", show=False, filename=r'c:/Users/dminkovski/Development/Microsoft/AzureDiagramMCP/diagrams/aks_landing_zone_architecture'):
    ad = ActiveDirectory("Azure AD (IdP)")
    management_rg = Cluster("Management Subscription")
    with management_rg:
        policy = Devops("Azure Policy")
        log_analytics = Logs("Log Analytics")
        monitor = Monitor("Azure Monitor")

    network = Cluster("Hub-and-Spoke Network")
    with network:
        hub_vnet = VirtualNetworks("Hub VNet")
        firewall = Firewall("Azure Firewall")
        bastion = Devops("Bastion / Jumpbox")
        with Cluster("Spoke: AKS VNet"):
            aks_subnet = Subnets("AKS Subnet (Private)")
            aks = AKS("AKS - Private Cluster")
            app_gw = ApplicationGateway("Application Gateway (Ingress)")
            lb = LoadBalancers("Internal Load Balancer")

    # Supporting services
    acr = ContainerRegistries("ACR")
    keyvault = KeyVaults("Key Vault")
    storage = StorageAccounts("Storage Account")
    sql = SQLManagedInstances("Azure SQL MI")
    cosmos = CosmosDb("Cosmos DB")

    # CI/CD
    cicd = Pipelines("CI/CD (GitHub Actions / Azure Pipelines)")

    # Connectivity and flows
    ad >> aks
    cicd >> Edge(label="push images") >> acr
    acr >> Edge(label="pull images") >> aks
    app_gw >> lb >> aks

    aks >> log_analytics
    aks >> monitor
    aks >> keyvault
    aks >> storage
    aks >> sql
    aks >> cosmos

    firewall >> hub_vnet
    hub_vnet >> Edge(label="peering") >> aks_subnet
    bastion >> Edge(label="admin") >> aks
    policy >> Edge(label="governance") >> aks
    log_analytics >> monitor

Common Mistakes or Myths

Let’s debunk some nonsense:

❌ “The AI diagrams won’t be accurate.”
Actually, they’re as accurate as your description. And since the output is code, you can review & tweak.
❌ “Non-technical stakeholders won’t get it.”
Wrong. Non-tech folks will love it because they can just say what they want in plain English and get a diagram.
❌ “We’ll lose control if AI does it.”
Nope. You’re still the architect. AI is just drawing faster.

⚠️ Some Mistakes to avoid:

Overloading diagrams with detail.
Nobody cares about every subnet. You can refine that yourself.
Show the big picture. Oh … you a perfectionist? Sorry to hear.
Treating diagrams as static.
Static = dead. Automate refreshes or don’t bother.
Not versioning diagrams.
If it’s not in your git repo, it doesn’t exist.

Final Thoughts

We’re past the era of designing yourself.
The new era is AI diagramming.

With MCP + AI, you don’t draw diagrams anymore.
You talk them into existence. You iterate in real time.
You keep them alive in your repo.

And honestly? Once you try it, you will not want to ever go back.

So the next time you hear “Can you make me a diagram of this setup?” -don’t panic. Just use azure-diagram-mcp and type a prompt.

You’ll have a crisp diagram before they’ve even opened Visio.
Because seriously … ain’t nobody got time for Visio. 😅

👉 Your turn: Would you trust an AI-generated Azure diagram in a customer pitch… or would you still redraw it “by hand” for polish?