How to Host a Web App Built with Claude, ChatGPT or Cursor

What “built with Claude” actually gives you

“Built with Claude” can mean three very different things, and the hosting route depends on which one you have:

• A Claude Artifact (a single HTML or React file shown in the chat). It can be copied out and hosted as a static site, but features that rely on Claude’s own sandbox (such as built-in AI calls or storage) will not work outside it.
• Code pasted from chat (HTML/CSS/JS, Python, Node). This is plain source code. It runs anywhere that supports its language.
• A full project from Claude Code, Cursor, Lovable, Bolt or similar: a repository with a framework (Next.js, React + Vite, Flask, FastAPI, Express), a package file and often a database. This is a real application and needs real deployment practice.

The same applies to apps generated by ChatGPT, Gemini, Copilot or any other LLM: hosting is a normal software-deployment problem. The model does not host anything for you.

Step-by-step: from AI-generated code to a live URL

1. Get the code into Git. Create a GitHub or GitLab repository and commit everything except secrets. Add a .gitignore for node_modules, .env and build output.
2. Run it locally. Install dependencies and start it (npm install && npm run dev, or pip install -r requirements.txt). If it does not run locally it will not run in production.
3. Find every secret. Search the code for API keys, database URLs and tokens. AI-generated code sometimes hard-codes them. Move them to environment variables and rotate any key that was ever pasted into a chat or committed.
4. Pick the hosting path using the table below: static host, serverless platform, container platform or VPS.
5. Provision the database if there is one (managed Postgres such as Neon or Supabase, or MongoDB Atlas). Never ship SQLite files on ephemeral serverless disks.
6. Set environment variables in the host’s dashboard, then trigger a build. Fix build errors (usually Node version, missing dependency or case-sensitive file paths).
7. Connect your domain. Add the A/CNAME records your host gives you. HTTPS certificates are issued automatically on all major platforms.
8. Turn on CI/CD. Connect the repo so every merge to main deploys automatically, with preview deployments for pull requests.
9. Add monitoring and backups. Uptime checks, error tracking (Sentry or similar), database backups and a spending cap on any paid AI API.
10. Document it. One page: where it is hosted, who owns the accounts, how to redeploy, how to restore. This is the step most AI-built projects skip.

Which hosting should you choose?

*Indicative ranges at the time of writing; provider pricing changes, so check current plans.

The seven mistakes AI-built apps ship with

• API keys in front-end code. Anything in the browser bundle is public. Call paid APIs (OpenAI, Anthropic, Stripe) from a server only.
• No server-side authorisation. Hiding a button is not security. Every API route must check who is calling.
• Open database rules. With Supabase or Firebase, forgetting row-level security or rules exposes all data to anyone with the public key.
• No rate limiting. A public endpoint that calls an LLM can be drained of credit in hours.
• Dev server in production. Run the production build behind a proper process manager or platform.
• No error handling or logging. You cannot fix what you cannot see.
• Nobody owns the accounts. The app lives in a freelancer’s or founder’s personal cloud account with no recovery path.

See our production checklist for AI-built apps for the full list, or hire a rescue review if the app is already live and misbehaving.

Making the hosted app visible in Google and AI search

Single-page apps that render everything in the browser are hard for crawlers, and many AI crawlers do not execute JavaScript at all. If you want the app’s public pages to rank, use server-side rendering or pre-rendering, give every page a unique title and description, and follow our guide to making a website LLM-friendly.

How AWI Digital helps

We deploy and support apps built with Claude, ChatGPT, Cursor, Lovable, Bolt, v0 and Replit for startups and SMEs. Typical engagements: take your repo live with CI/CD, domain and HTTPS; migrate off a prototype platform; harden security; add monitoring; or run ongoing support. Fast website launches start at $250, and quotes for apps are scoped after a short code review. Request a free review.

Frequently asked questions

Related guides