Building Your Own AI Development Environment with ChatGPT and Claude

Building Your Own AI Development Environment with ChatGPT and Claude
Using AI like ChatGPT or Claude to write code has become commonplace in recent times.
Previously, developers mostly asked questions about source code and copied answers directly into files. Now, AI can read project files, understand file relationships, modify code, and even run tests within the development environment.
I've been using a similar approach—copying relevant source code to ask AI for modifications and applying the changes back to my server. While this works for simple fixes, it becomes cumbersome as projects grow larger.
With multiple files, deciding which ones to share with AI gets complicated. Long code snippets are hard to copy selectively, and there's a risk of accidentally breaking existing functionality when applying modified code.
So this time, I'm moving beyond simply asking AI for code suggestions. Instead, I'll create an environment where AI can work directly within the development setup.
The Development Environment We're Building
This project aims to set up a dedicated virtual machine on a Proxmox server with Git, Docker, Claude Code, and MCP connected.
Here's how everything fits together:
My PC
├── ChatGPT
├── Claude
└── Web Browser
Proxmox Server
└── AI Development VM
├── Docker
├── Forgejo Git Server
├── Claude Code
├── MCP Server
├── Git Workspace
└── Test Environment
Production Server
├── PHP Project
├── Node.js Project
├── MySQL
├── Nginx
└── PM2
The AI development VM will be separate from the production server.
Instead of applying changes directly to live sites, modifications happen in a dedicated workspace. After testing, changes are committed via Git and then reflected on the production server.
Why Use Both ChatGPT and Claude?
While ChatGPT and Claude seem similar, they excel at different tasks in practice.
I plan to divide responsibilities rather than compete between them.
ChatGPT will handle overall project structure design, feature planning, error analysis, and content creation like blog posts and image generation.
Claude Code will focus on actual code modifications within the project files, testing, and development tasks.
For example, when adding a new feature:
ChatGPT
→ Design feature structure and database schema
Claude Code
→ Modify actual project files
Git
→ Save changes
Test Server
→ Verify functionality
Production Server
→ Final deployment
This division allows AI to function as an active participant in the development process rather than just a question-answering tool.
Why Set Up a Git Server?
When people think of Git, GitHub often comes to mind first.
However, Git doesn't require external services. You can run it entirely on your own server, even within Proxmox as a private Git server.
This project will use Forgejo to set up a personal Git server.
Source code stays internal and only change history is tracked systematically.
With Git, if Claude makes an error, you can revert to previous versions easily.
Save pre-modification state
→ Claude modifies code
→ Compare changes
→ Issue occurs
→ Revert to previous version
The more AI participates in development, the more essential Git becomes. Just like humans, AI makes mistakes.
Why Use Docker?
You could set up Forgejo and MCP servers through traditional installation methods.
But installing multiple programs directly can lead to package version conflicts or complex configuration files. Rebuilding the same environment later would be difficult.
Docker allows each service to run in isolated containers:
Docker
├── Forgejo Container
├── MCP Container
└── Additional required services
If an issue arises with one service, you can restart just that container. Backing up configuration files and data folders makes migrating to another server straightforward.
This project will run services via Docker while installing tools like Claude Code and Git directly on the VM.
What Role Does MCP Play?
MCP serves as a bridge connecting AI with external tools.
Claude needs this connection to do more than just modify files—it needs to query databases, check Git status, or read server logs.
With MCP connected, Claude can use various tools within defined permissions.
Examples include:
Claude
├── Read project files
├── Modify source code
├── Check Git changes
├── Query MySQL database
├── Review server logs
├── Test via browser
└── Execute deployment scripts
I won't grant all permissions upfront.
For example, the production database will connect through a read-only account. Dangerous actions like file deletion or mass modifications will be restricted.
It's safer to gradually grant only necessary permissions rather than giving AI broad access.
Why Separate Development and Production Servers?
You could install Claude Code directly on your production server and modify code there.
But this risks causing immediate service outages from small mistakes. For example, a single PHP file modification might break the entire site, or incorrect Node.js configuration could stop PM2 processes.
So this project separates development and production environments:
Development VM
→ Copy code
→ Claude modifies
→ Test
→ Commit to Git
→ Reflect on production
This structure allows for thorough review before applying changes to the live server.
What We'll Actually Do in This Series
This isn't just a theoretical article. I'll actually set up a VM on Proxmox, install Ubuntu Server, and configure Docker and Forgejo.
I'll then connect Claude Code and MCP, bring existing PHP and Node.js projects into the test environment, and perform actual code modifications.
The process will follow these steps:
Creating an AI Development VM on Proxmox
Installing Ubuntu Server and configuring CPU, memory, disk, and network settings.
Installing Docker and Docker Compose
Setting up the base environment for running Forgejo and MCP servers.
Setting Up a Private Git Server with Forgejo
Creating an internal-only Git repository accessible within your private network.
Managing Existing Projects in Git
Registering current PHP and Node.js projects in the Git repository and saving their initial version.
Installing Claude Code
Setting up Claude Code on the development VM to read project folders.
Connecting the MCP Server
Gradually connecting file systems, Git, MySQL, SSH, and browser automation tools.
Performing Actual Project Modifications
Starting with simple feature modifications, tracking changes via Git, and testing them.
Backing Up the Development Environment
Creating a secure backup structure for Git data, Docker containers, VMs, and databases.
The Final Workflow We're Aiming For
Once completed, this setup will enable:
You
→ Request feature
ChatGPT
→ Plan features and design structure
Claude Code
→ Analyze project and modify code
Git
→ Record changes
Test Environment
→ Check for errors
You
→ Final review
Production Server
→ Safe deployment
Instead of manually copying AI-generated code, the goal is to have AI work like a developer within the environment.
I won't fully automate everything. Important changes will still require human verification and approval.
Closing Thoughts
This project goes beyond simply installing Claude Code. It connects Proxmox, Docker, Forgejo, Git, and MCP into one cohesive development flow, applying it to actual production projects.
I'll simplify the setup process and explain configuration reasoning so even those without deep technical knowledge can follow along.
In the next installment, I'll start by creating an AI development VM on Proxmox and installing Ubuntu Server.
FAQ
Why separate development and production environments?
To prevent accidental service disruptions from AI modifications; changes are tested in a safe environment before being deployed to live servers.
What is MCP and why is it important?
MCP connects AI with external tools like databases and Git, allowing Claude Code to perform tasks beyond simple code editing within defined security limits.
Korean Source
This article is an English localization of a Korean-language report from BlogScore. Original: [AI 개발환경 구축 프로젝트] 프롤로그. ChatGPT와 Claude를 함께 쓰는 나만의 AI 개발 환경 구축 프로젝트