You asked ChatGPT or Claude for a small tool, and it delivered. Two hundred lines of code, a friendly explanation, and a last line that says something like "save this as index.html and deploy it to your preferred hosting provider."
That last line is where most people stop. The tool exists, in a sense, but only as text in a chat. Nobody on your team can open it, and you are not sure what "deploy" involves or whether you are about to break something.
This guide is for that moment. It tells you how to work out what kind of code you have, the quickest honest way to put each kind online, and how to skip the step entirely next time by letting the assistant finish the job.
Quick answer
It depends on what the code does.
- A single page that only shows things (a landing page, a calculator that runs in the browser, a menu): drag the folder onto a static host such as Netlify Drop. You get a public link in about a minute, and Netlify lets you do it before you even create an account.
- Anything that stores data, sends email, runs on a schedule, has a login, or uses an API key: that needs a server, a computer that keeps running when your laptop is closed. A static host cannot run it, however you upload it.
- Either way, the easiest route next time is to connect a builder to the chat app you already use. Server4Agent is one: the assistant writes the code on a real server, starts it, and replies with a private link instead of a block of text.
First, work out what you have
You do not need to read the code. Look at the file names the assistant gave you and at a few words in its explanation.
| If you see | You probably have | Where it can live |
|---|---|---|
Only index.html, maybe style.css and script.js | A static page | Any static host, in a minute |
package.json, server.js, app.py, requirements.txt | An app with a server part | A server |
| "database", "SQLite", "store submissions", "save to a file" | Something that keeps data | A server with storage that persists |
| "cron", "every morning", "schedule", "runs daily" | A scheduled job | A server that runs on its own |
"API key", "token", ".env" | Something that talks to another service with a secret | A server, where the secret stays hidden |
If any row below the first one matches, a drag-and-drop host is the wrong tool, even if it accepts your upload. The page may load, but the parts that matter will quietly not work.
The one-minute route: a static host
For a page that only shows things, a static host is the right answer and it is free.
- Put the files the assistant gave you into one folder, with the main file named
index.html. - Open Netlify Drop and drag the folder onto the page.
- You get a link on
netlify.app. Claim the site with a free account if you want to keep it and update it later.
Vercel Drop works the same way and also takes unbuilt projects, but it asks you to sign in first. GitHub Pages is free too, if you already have a GitHub account and do not mind learning what a repository is.
Two warnings before you publish:
- Never put an API key in a static page. Everything in it is sent to every visitor's browser, so anyone can read the key and spend your money with it. If the assistant's code has a line like
const API_KEY = "sk-...", the code needs a server. - A form on a static page sends nowhere by default. If people should be able to submit something, the submissions need somewhere to go, which is a server or a third-party form service.
The other quick routes, and where they stop
A preview inside the chat app. Some chat apps can show a simple page right in the conversation, and some can share it as a link. That is great for checking an idea. It is not a tool your team can rely on: it keeps no data of its own and does not run anything after the chat is closed.
An app builder. Products built around their own editor can host what you build in them. The catch is that you start over inside their product, so the conversation you already had, and the assistant you already trust, stay behind.
A developer. A developer can set up hosting for anything in an afternoon. For a small internal tool, that is usually the most expensive afternoon in the whole project, and you are back to asking someone else every time it needs a change.
The route that skips the step: let the assistant finish
The reason the assistant hands you code and stops is simple: on its own, it has nowhere to run anything. Give it a server and the last step disappears.
Server4Agent is a hosted server for AI assistants. You add it once to the chat app you already use, and from then on "put this online" is a request the chat can finish:
- ChatGPT: add Server4Agent as a custom MCP server on the Plugins page and sign in. The ChatGPT setup guide walks through each screen.
- Claude: add it as a custom connector from the connectors menu and sign in.
- Cursor, VS Code and other assistants that support MCP: the integrations page has the steps for each.
Then say what you want in plain words. Here is how it plays out for the office manager at Northstar Labs, who had an equipment booking page from an earlier chat:
> "Put the booking page you wrote yesterday online for my team. Bookings should be saved, and I want an email when someone books the projector."
The assistant creates a project on a server, writes the files, installs what the code needs, starts it and checks that it answers. It asks who should be able to open it. The office manager picks a password the team can share, and the assistant replies with a link on apps.server4agent.com. Bookings are stored on the server, the email goes out when someone books, and all of it keeps working after the chat is closed. When the office manager writes back "only show weekdays", the same link updates.
Two differences from every route above matter more than the convenience:
- The assistant can see whether its code actually runs. In a plain chat, it writes code it has never executed, and you find the bugs. On a server, it runs the code, reads the error, and fixes it before you see anything.
- Nothing is public until you decide. A new project serves nothing until you choose who can open it: your team only, anyone with a password, or everyone. The address stays the same whichever you pick.
What "online" should mean for a tool your team uses
Whichever route you choose, check the result against this list before you send the link around:
- The link works when your laptop is closed.
- Data people enter is still there next week.
- It is private until you choose who can open it.
- Keys and passwords live on the server, not in the page.
- Spending has a ceiling. On Server4Agent that is a budget cap on the account, and the assistant cannot raise it.
- Someone else can take it over. Every Server4Agent project is a git repository an engineer can open in a browser editor.
A static page passes the first point and maybe the third. Anything more than a page needs a server to pass the rest.
FAQ
Do I need to learn to code to do this?
No. For a static page you need to drag a folder. For anything bigger, connecting a builder to the chat means you describe the result and review the link, the same way you would brief a colleague. The non-technical guide to shipping tools from chat covers the habit.
The code the chat gave me does not work. Is it my fault?
Almost never. An assistant without a server writes code it cannot run, so small mistakes are normal. Paste the error back into the chat, or give the assistant a server so it can run the code and catch the error itself.
Is a free static host enough for an internal tool?
Only if the tool is a page that shows information and stores nothing. The moment it needs to save a form, send a message, or keep a secret, it needs a server. The first tool to ask your AI assistant to build suggests starting points that fit either case.
What does Server4Agent cost?
There is a free plan with one small server, one live public project and a little trial AI credit. Starter is $29 a month for two servers and three live projects. Full limits are on the pricing page, and a budget cap you set keeps the total bounded.
Can I move the tool somewhere else later?
Yes. Each project is a git repository with its full history, so an engineer can export it and run it anywhere else.
Related reading
- The non-technical guide to shipping real tools from chat
- Why every AI agent should be able to return a URL
- Share an internal tool with your team without putting it on the internet
- How to give your AI assistant a real computer