thelastapi
Get started free

Deploy a website from Claude or ChatGPT with nothing installed

Deploy a website directly from ChatGPT with nothing installed by connecting the remote MCP connector and asking ChatGPT to build and publish it.

On this page 6 sections

You can deploy a website directly from ChatGPT with nothing installed on your computer by connecting thelastapi as a remote MCP connector. In ChatGPT settings, you turn on Developer mode, add the connector URL https://thelastapi.dev/mcp, and authorize your account through standard OAuth in your browser. From that point on, you can tell ChatGPT to create, style, and publish a website, and it will deploy the files directly to a live <slug>.vibiz.app address.

This guide walks through setting up the remote connector, prompting ChatGPT to deploy sites, making live edits, handling secrets safely, and staying within chat limits.

How ChatGPT deploys websites without local tools

Deploying a website traditionally requires installing a command-line tool, downloading runtimes like Node.js, cloning a repository, and wiring up authentication keys on your local machine. If you are on an iPad, a work computer with locked permissions, or your phone, that workflow breaks immediately.

ChatGPT solves this by using the Model Context Protocol (MCP) over HTTP. Instead of running a local server on your machine, ChatGPT connects directly to a remote MCP endpoint hosted at:

Text
https://thelastapi.dev/mcp

When you add this URL to ChatGPT, the model discovers the deployment tools exposed by thelastapi. The authentication runs over OAuth 2.1 with PKCE. The first time ChatGPT attempts to interact with your hosting or resources, it directs you to an Authorize page in your browser. You sign in with your thelastapi account and grant access.

Once approved, ChatGPT receives a dedicated API key listed on your API keys page under the client's name. You never have to copy, paste, or expose secrets inside the chat window. Everything happens server to server. If you want to connect Claude, Cursor, or Codex to the same setup, read our guide on how to connect a remote MCP server to Claude, ChatGPT, Cursor and Codex.

How to connect the remote connector in ChatGPT

Connecting ChatGPT takes under two minutes. You will need a ChatGPT plan that supports custom connectors and developer mode.

Step 1: Turn on Developer mode

In ChatGPT, navigate to Settings ▸ Apps ▸ Advanced settings. Toggle Developer mode on so ChatGPT can connect to external servers.

Step 2: Add the connector

Under your app connector settings, create a new connection using the remote URL:

Text
https://thelastapi.dev/mcp

Configure it with OAuth sign-in. Name the connection Vibiz or thelastapi.

Step 3: Authorize access

When prompted, open the authorization link in your browser. You will see a permission screen with two access levels:

Access levelPermissionsSafe for
Full access (default)Can deploy, create projects, manage secrets, configure domains, handle live payments, and delete resources.Trusted automations and private workspaces.
LimitedCan deploy sites, run actions, send email, use AI, and create projects. Blocked from live money and production secrets.Daily chats, experimental agents, and public prompts.

Select your preferred access level and click Authorize. If your account is already set up, ChatGPT connects instantly. While new sign-ups are closed, anyone signing up without an account will receive reason: signup_closed with a link to follow instead.

For detailed permission breakdowns, check the documentation on Keys and environments and Agents.

Prompting ChatGPT to build and publish your site

Once connected, ChatGPT can access hosting tools including vibiz_create_project and vibiz_deploy. You do not need to instruct ChatGPT on which technical endpoints to call. You can describe what you want in plain English.

Here is a prompt you can paste directly into a new chat:

Text
Use thelastapi to build and publish a landing page for my bakery, Acme Bakery, with a contact form.

When you submit this prompt, ChatGPT carries out two tasks automatically:

  1. Creates a project: It calls vibiz_create_project. The platform provisions the required hosting resources along with a unique slug.
  2. Deploys the files: ChatGPT writes the HTML, CSS, client scripts, and assets, packaging them into an array of file paths and contents (files: [{ path, content }]). It then calls vibiz_deploy with the new project ID and your files.

ChatGPT returns the live URL where your site is instantly reachable:

Text
https://acme-bakery.vibiz.app

HTTPS certificates are provisioned automatically. You can read more about how AI models structure complete web pages in our walkthrough on how to host a landing page that ChatGPT or Claude wrote.

Updating pages, checking code, and rolling back

After your initial deployment, you can continue iterating on the website without ever opening a code editor. ChatGPT manages edits through the project inspection tools.

Editing existing files

When you ask ChatGPT to make a change, such as:

Text
Change the hero background color to warm cream and add an opening hours section below the menu.

ChatGPT calls vibiz_code to inspect the live project files. It reads the files currently serving on your site, identifies the necessary modifications, and calls vibiz_deploy passing only the changed files. Every deployment includes an audit message in your version history.

Rolling back a bad update

If ChatGPT introduces a visual bug or breaks your layout, you do not have to debug the code inside the chat. You can simply ask:

Text
Roll back the site to the previous deployment.

ChatGPT invokes vibiz_rollback, which points the live domain back to the preceding deployment version. Keep in mind that while website code rolls back instantly, the database is never rolled back automatically, preventing unintended data loss.

Supplying secrets without revealing them in chat

If your website requires a third-party token or external API credential, never paste the raw key into your chat dialogue. When a key is needed, ChatGPT runs vibiz_request_secret. This tool generates a private console link:

Text
Paste your secret key here: https://thelastapi.dev/projects/prj_123/secrets/new?name=API_KEY&env=production&kind=env. It goes straight to thelastapi, never through this chat.

You open the URL, paste your secret securely into the encrypted console form, and the environment variable is injected directly into your project's hosting environment.

Chat limits and moving to your local machine

Deploying websites through a chat interface is fast, but browsers and chat context windows have natural limits.

ParameterLimit in ChatGPT / Claude chatSolution when exceeded
Max files per deployUp to 300 filesMove to a local coding agent or CLI
File size limit3 MiB per fileHost media in project Files
Total deployment payload4 MiB combinedPull project locally with vibiz pull
Local file system accessUnavailableUse local MCP server (vibiz mcp)

When your application outgrows a simple landing page or lightweight app, you can pull the entire project onto your computer using the command-line interface.

Install the CLI via npm and log in:

Terminal
npx @vibiz/cli login

Then download your live code:

Terminal
mkdir acme-bakery && cd acme-bakery
npx @vibiz/cli pull

This writes your live deployment files into your local directory and places a vibiz.json file in your root folder. From there, you can continue working locally, run vibiz dev to test against a development database branch, or hook up coding agents like Claude Code, Codex, or Cursor using our guide on how to configure one API key for an AI agent: hosting, database, email, payments and models. Both the chat interface and your terminal share the same account, code history, and hosting platform.

To set up an account, review deployment options, or learn how to manage database resources alongside hosting, visit the Quickstart guide and create your account at thelastapi.dev/start.

Frequently asked questions

Do I need Node.js or npm installed on my machine to deploy from ChatGPT?

No, you do not need any software installed. ChatGPT connects directly to the remote MCP server at `https://thelastapi.dev/mcp` over HTTP and deploys your files via OAuth.

Can ChatGPT deploy to a custom domain I own?

Yes, you can manage domains using the built-in domain tools, though purchasing or adding a domain from chat requires your key to have the necessary permissions.

What happens if ChatGPT makes a mistake and breaks my live site?

You can tell ChatGPT to roll back the change immediately, which executes `vibiz_rollback` to return to your prior working deployment version.

What are the payload limits when deploying through a chat connector?

Each deploy from a chat app carries up to 300 files, with a limit of 3 MiB per file and a total payload limit of 4 MiB.