It’s a sleek, powerful preview app that you will love.
If you are not a premium member, read the full article here and consider supporting my work here on Medium and on YouTube.
I’ve spent the last few days testing it across different projects, and I’ve really come to like it.
But there’s a glitch I discovered after building three projects back-to-back that you need to know about.
Here’s what makes the desktop version different from the terminal workflow you’re used to.
The terminal version requires you to jump between your code editor and command line, typing prompts and managing files manually. The desktop app is a dedicated interface with built-in folder selection, environment controls, and model switching.
The interface is a new tab tacked to your Claude Desktop app sidebar.
Click the code tab, and you have everything in one place.
Select your project folder from the dropdown, choose your environment (local by default), pick your model from Sonnet 4.5 or others, and start prompting.
Three prompt templates sit at the bottom to help you get started quickly. One creates Claude.md files for project documentation.
Another searches your codebase for TODO comments and fixes them. The third recommends areas to improve or test in your project.
This is faster than switching contexts in your terminal because you can see the file structure, watch code being written in real-time, and open VS Code directly from the interface with one click.
I’ll show you what works, where it breaks, and what you can do about it.
We’ll start with the basics of setting up the interface, then move into building a complete full-stack project that demonstrates its core features.
By the end of this article, you’ll know everything you need to use Claude Code Desktop and leverage it to build faster than you can imagine.
I shared the same on my YouTube channel:
But this article covers everything you need to know, along with live-cast demos that will quickly help you learn how to use it.
Getting Started with Claude Code Desktop
You need to download and install the Claude Desktop app first. Claude Code lives inside it.
Once installed, look for the code tab in the sidebar menu.
Click it, and you’ll see the welcome screen prompting you to install runtime dependencies. Click install dependencies to begin the setup.
Claude Code Desktop User Interface
After installation completes, you get the main Claude Code desktop interface.
- The prompt input sits at the center.
- Above it, you have three key controls.
- Select folder option for choosing your project directory.
- Environment selection dropdown set to local by default
- Model selection with the latest available models, like Claude Sonnet 4.5.
Below the prompt input, three starter templates help you create quick prompts.
- Create or update Claude.md file.
- Search for TODO comments and fix them.
- Recommend areas to improve or test.
At the bottom, a warning notice about potential security risks when Claude Code accesses your file system.
Setting Up Your Environment
Click the folder select option to choose your project folder.
The environment dropdown lets you work locally or add custom cloud environments by filling in access details and permissions.
The model selection gives you access to the latest Claude models. Pick what you need for your project.
When you first try to access a workspace, a permissions popup appears. Click Trust workspace to confirm and continue. This happens in VS Code too when Claude Code opens files there.
You’re now ready to start building with Claude Code desktop.
Core Features
Let me show you how these prompt templates work.
Creating Your First Prompt
Click the first prompt template to insert it in the prompt input. Select your folder, environment, and model, then send the prompt.
The trust workspace warning pops up immediately. Click the Trust workspace to continue.
Claude Code starts creating the Claude.md file inside your selected project folder. At the bottom, a button appears to open this in VS Code.
When completed, you see the code inside the artifact along with a summary of what was created. The file sits ready in your project folder.
Click open in VS Code. Another permission notification pops up requiring you to confirm before proceeding. The same happens in VS Code where you just confirm to continue.
Inside, you see the sample Claude.md file. It’s a basic template you can customize to fit your project needs.
Searching Your Codebase
Claude Code desktop has an efficient search function built in. I added a TODO comment to my Claude.MD file to demonstrate this.
Back in the Claude Code interface, I clicked the second prompt template to search my codebase. Select the model and send the prompt.
It immediately starts searching and asks for permissions.
Then it returns with a message that this is an empty directory with just a Claude.md file. It caught the TODO in the file and wants to remove it.
It got the TODO with a preview of the code.
I confirmed that it should remove it. Opening the VS Code file shows it successfully removed the TODO and gave feedback that the issue was fixed.
This is how you fix pending issues in your large codebase using the search-and-fix approach. Claude Code finds the TODO pending issues and fixes them.
Testing Project: Building ASE Code Time Tracker
Time for the real coding test. I decided to build a coding time tracker to see how Claude Code desktop handles a complete full-stack project.
Here’s the prompt I used:
I want you to help me build a simple full-stack web app called
“ASE Code Time Tracker.”
It should track my daily coding hours. Keep the architecture extremely
simple—no unnecessary complexity—but make the UI clean and pleasant.
Tech Stack
Backend: Python FastAPI
Frontend: Vanilla JavaScript, HTML, CSS
Database: SQLite (or the simplest built-in option)
Core Requirements
Homepage with:
Start button to begin timing a coding session
Stop button to end the session
A live timer display
Auto-submit of session data to the backend
Backend endpoints:
POST /sessions/start — record session start
POST /sessions/stop — record stop and duration
GET /sessions/today — return total hours coded today
GET /sessions/all — list all past sessions
Data Model:
id
start_time
end_time
duration_seconds
Frontend:
Use fetch() to call the API
Display today’s total coding time
Auto-refresh after each session
Styling:
Minimal CSS
Clean, modern, simple layout
Fully vanilla — no frameworks
Delivery:
Provide full code for:
main.py (FastAPI)
HTML, CSS, JS
SQLite setup
Provide instructions for running everything
Extra Notes
Keep code beginner-friendly with clear comments.
Avoid over-engineering.
I put the prompt in the input, selected my project folder, left the environment as local, chose Claude Sonnet 4.5, and clicked send.
Building Process
Claude immediately created a to-do list summarizing the project into seven actionable steps:
- Create FastAPI backend
- Create database initialization
- Create HTML frontend
- Create CSS styling
- Create JavaScript logic
- Create requirements.txt
- Create README with setup instructions
I opened VS Code side by side with Claude Code. This helps me follow the process as files get created and code gets written.
Claude started with the backend. Created main.py, complete with all the code.
Then, it moved to the database initialization script. On the task list, you can see it checks off every completed task.
It moved to the frontend and completed index.html. Then worked on the styles. Then, JavaScript logic was used to write the timer code and all the functionality we needed.
Finally, it wrote the README with all the setup instructions.
Problem I Discovered
This is where I hit my first issue with the Claude Code desktop.
I gave it a prompt to open the app in a live server and run it from the terminal. It froze for nearly three minutes.
I quickly discovered that running this app is memory-intensive. It freezes at times, and the only way out is to close the session and open it fresh. You need to exit it completely from the minimized programs tray.
I started running it as admin to give the program full privileges. This helped with the freezing issue.
The app works well once it’s running. It opened the live preview, and everything functioned as expected.
I used the last prompt template to get recommendations for areas to improve or test. It gave a detailed list of all possible tests we could run to make the project production-ready.
The desktop version delivers on speed and workflow integration. But you need to know about the memory issue before using it in large projects.
