Deploying an app built with Claude Code: GitHub, servers, domains and backups

Published on 2026-10-27 by frfrand

Steps to turn Claude Code code into production software: CI/CD, systemd, secrets, VPS, testing, and rollback.

Deploying an app built with Claude Code: GitHub, servers, domains and backups

The difference between 'runs on my terminal' and 'works tomorrow'

Claude Code can create and modify a repository and even run tests where it has permissions. A business application needs the result not to depend on an open session or on an assistant remembering how to restart a process. That requires documented deployment and repeatable operations.

Three limits worth setting

First, that changes remain versioned in Git and are reviewed. Second, that an agent does not have by default secrets for all services nor root permissions if it only needs to modify the frontend. Third, that there is a staging environment with test data to validate modifications before touching production. Keys and credentials must be stored per service, with the minimum possible scope.

Example: course portal generated with Claude

The frontend allows choosing a class and a backend saves enrollments. The demo works, but upon restart it disappears because the Node process was launched manually. You configure a systemd service, a persistent PostgreSQL database, and Nginx with HTTPS. You test a complete enrollment, an email error, and a backup restoration. In a subsequent deployment the login fails: the previous commit and the rollback process allow recovering functionality before investigating.

Concrete sequence

1. Inventory the repository's processes and dependencies. 2. Choose a maintained runtime version and register it in the project. 3. Prepare CI to run build and minimal tests. 4. Configure persistent service, proxy, certificate, and secure variables. 5. Restore a backup in staging to demonstrate recovery. 6. Publish from Git and monitor errors and latency.

What ArduMaker can do in that journey

The infrastructure can host runtime, domain, and services, with deploy operations as needed. The important thing is that the project remains yours and there is a procedure for another person to maintain it. There is no need to stop using Claude: it serves as a development tool within a controlled flow.

Frequently asked questions

### Can I give root access to an agent? Avoid unnecessary privileges; define permissions and approvals. ### Can the app use Docker Compose? Yes, if it helps coordinate services; it requires volumes, backups, and networks correctly configured. ### What is a rollback? Returning in a controlled manner to a previous release when the new one fails. ### Who fixes errors after delivery? Define it in writing in the support and maintenance scope.

Sources

GitHub Actions · Docker Compose Production · Nginx Docs.

Moving from AI code to ArduMaker migration service

The goal of the future service at ardumaker.com/lovable is to study AI-generated code, adapt it to a working deployment, move services if agreed, and publish it on controlled infrastructure. The page does not exist yet and we are not announcing an available automatic migration tool. At ArduMaker you can check the current VPS and services infrastructure. Each project requires diagnosis: a static landing page, an SSR with database, and a SaaS that charges subscriptions do not need the same process.

Editorial note: October 2026. Test first on staging, document licenses, and keep verified copies before modifying production.

Deploying an app built with Claude Code: GitHub, servers, domains and backups

What the photograph shows

Checking the status of a mobile application. Illustrative photograph generated to accompany this guide on migrating applications created with AI; it is not a screenshot of the platform nor does it certify a specific installation.

Continue with these practical guides from ArduMaker

- Taking Cursor-created code to production: from editor to an app that doesn't crash - From ChatGPT or Codex to your own website: publishing AI code in production - Deploying AI-created Next.js on a VPS: SSR, API Routes and Node.js

If you are evaluating your own application, compare the real backend, database, authorization and deployment requirements before choosing the server.

  • Claude Code
  • IA
  • deploy
  • GitHub
  • infraestructura

On ArduMaker

  • Deploy from GitHub: Link a repo and every push to your chosen branch updates your app on the VPS, with build and restart scripts.
  • Automatic backups: A full live backup every week, the last 4 kept, and a final copy when you delete the VPS.

Keep reading

All articles · VPS plans