How to build a website with VS Code, Gemini CLI, Antigravity or any MCP client

Add PageHub to your AI tool once, then describe the site you want. Your agent builds it, checks it with screenshots, and puts it on your domain. You end up with a real site that anyone can edit visually, not a folder of code you still have to host.

What you need: An MCP client such as VS Code with GitHub Copilot, Gemini CLI or Antigravity, a free PageHub account, and a domain if you want one. Without a domain, the site publishes to a pagehub.dev subdomain.

How it works

PageHub runs an MCP server, which is a set of tools an AI agent can call. VS Code, Gemini CLI, Antigravity and most other MCP clients support remote MCP servers over HTTP with OAuth sign-in, so once you add PageHub, your agent can call tools like create_site, set_theme, screenshot_site, and publish_site for you.

The site lives in PageHub, not in your project folder. PageHub also sends its own build rules to your agent when it connects, so you don't paste any instructions.

Step 1: Create a free PageHub account

Go to pagehub.dev/signin and enter your email. You'll get a magic link and a 6-digit code. Use either one to sign in.

The free plan has no credit card and no time limit. It includes 1 published site, 5 pages per site, and hosting with SSL, and you can connect 1 domain you own. Connecting your MCP client works on every plan.

Step 2: Add the PageHub MCP server to your MCP client

Every client connects to the same server, https://pagehub.dev/api/mcp, and signs you in through your browser. There's no API key to paste. What changes is where the setting goes:

ClientWhere PageHub goesHow you sign in
VS Code (Copilot).vscode/mcp.json or your user mcp.json, under servers with "type": "http"Click Start above the entry
Gemini CLIOne command, saved to ~/.gemini/settings.json/mcp auth PageHub
Antigravity~/.gemini/config/mcp_config.json, as serverUrl/mcp opens the MCP Manager
Any other clientIts MCP settings, as a remote HTTP serverWhen it opens PageHub in your browser

In each case the browser sign-in is the same: sign in with your email, then click Approve on the Authorize PageHub screen. Using Claude Code or Cursor? They have their own guides: Claude Code and Cursor.

VS Code and Copilot

The quick way: click the button. Your browser hands off to VS Code, which asks you to install PageHub.

Install in VS Code

To add it by hand, run MCP: Add Server from the Command Palette, or put this in .vscode/mcp.json (one project) or your user mcp.json (every project; run MCP: Open User Configuration):

mcp.json
{
  "servers": {
    "PageHub": {
      "type": "http",
      "url": "https://pagehub.dev/api/mcp"
    }
  }
}

VS Code's file uses servers and "type": "http", not the mcpServers shape other apps use. Save it, click Start above the PageHub entry, trust the server, and sign in. To check it worked, open Copilot Chat, switch to Agent mode, and click Configure Tools. PageHub's tools should be listed. On a Copilot Business or Enterprise plan, your admin may need to allow MCP servers first.

Gemini CLI

Signed in with a Google account? Since June 18, 2026, Gemini CLI no longer works with a free, Google AI Pro, or Ultra sign-in. Google moved those users to Antigravity, so follow the Antigravity steps. With a paid Gemini API key or a company Gemini Code Assist license, run this in your terminal:

terminal
gemini mcp add --transport http --scope user PageHub https://pagehub.dev/api/mcp

--scope user saves PageHub to ~/.gemini/settings.json, so it's there in every folder. Without it, Gemini CLI saves it to .gemini/settings.json in the current folder only. Then start gemini and type /mcp auth PageHub to sign in. Type /mcp to check: PageHub should say CONNECTED. Gemini CLI asks before each PageHub tool unless you choose to always allow the server.

Antigravity

Open ~/.gemini/config/mcp_config.json, or create it, and add PageHub:

mcp_config.json
{
  "mcpServers": {
    "PageHub": {
      "serverUrl": "https://pagehub.dev/api/mcp"
    }
  }
}

Use serverUrl exactly. Antigravity doesn't accept url or httpUrl here. If the file already lists other servers, add the PageHub entry inside the mcpServers object you have. For one project only, put the same entry in .agents/mcp_config.json in that project. Then start agy (or open the app) and type /mcp. The MCP Manager asks you to sign in, and PageHub shows as active once you have.

Any other MCP client

Add PageHub as a remote server (often called HTTP or Streamable HTTP) with the URL https://pagehub.dev/api/mcp, then sign in when it opens PageHub in your browser. The MCP server docs have the config for more clients.

Whichever client you use, check it worked by asking: “List my PageHub sites.”

Step 3: Prompt your agent to build the site

Open your agent's chat (Copilot Chat in Agent mode, gemini or agy in a terminal) and describe the site. Say who it's for, which pages you want, and what visitors should do. Real details (hours, prices, your email) beat placeholders. Swap the bracketed parts in these prompts for your own.

A local business site

prompt
Use PageHub to build a website for [business name], a [type of business] in [city].
Pages: Home, Services (with prices), and Contact.
Hours: [days and times]. Address: [address]. Phone: [phone].
Add a contact form that asks for name, email, and what they need.
Keep the look [warm and simple / bold / minimal].
Screenshot it on desktop and mobile before you tell me it's done. Don't publish yet.

A portfolio

prompt
Use PageHub to create a portfolio site for me. I'm a [job title] in [city].
Home page: a short intro, then a grid of 6 projects, each with a title, a one-line summary, and an image.
Add an About page and a contact section with my email, [you@example.com].
Clean layout with lots of white space. Use stock photos for now and I'll swap in my own.
Show me screenshots before publishing.

A landing page for a side project

prompt
Use PageHub to make a one-page site for my side project, [name]: [one sentence on what it does].
Sections: a hero with an email sign-up form, 3 features, a short FAQ, and a footer.
Suggest 3 color palettes and let me pick one before you build.
When it's built, run an SEO check and fix what it finds. Don't publish until I say so.

From there, your agent looks up PageHub's templates and section blocks, creates the site with create_site, sets colors and fonts with set_theme, adds sections, and finds stock photos with find_image. Then it checks its work with screenshot_site.

Step 4: Review the draft and edit it yourself

Everything your agent builds lands in a draft. Nobody else sees it until you publish.

Ask for screenshots at any point. screenshot_site takes a real browser screenshot of the draft, one section at a time if you like, and at phone width (390 pixels). Your agent reads the screenshots and fixes what it sees. VS Code also shows them in the chat. To look for yourself, open the site in your PageHub dashboard.

To edit by hand, open your PageHub dashboard, pick the site, and click Design. That opens the visual editor. Drag sections around, change colors and fonts, and rewrite text. your agent reads the same draft, so it sees your edits the next time you ask for changes.

Before you publish, ask your agent to run audit_seo and audit_accessibility and fix what they find.

Step 5: Publish and connect your domain

Ask your agent: “Connect [example.com] to this site and tell me the DNS records to add.” It calls set_domain, which attaches the domain and returns the records. Add them where you bought the domain:

  • The bare domain (example.com) gets an A record at @.
  • A subdomain such as www gets a CNAME record.
  • Sometimes there's also a TXT record, to prove you own the domain.

Copy each value exactly as your agent gives it to you. The values are specific to your site, so a value from another guide won't work.

PageHub sets up both example.com and www.example.com, and by default sends www visitors to the bare domain. SSL certificates are issued automatically and can take a few minutes once DNS is in place.

Then ask: “Is my domain working yet?” your agent checks with get_domain_status. Connecting a domain doesn't publish the site, so finish with: “Publish the site.” That calls publish_site, and the site is live.

Rather click? The Domains page for your site in the PageHub dashboard shows the same records.

Step 6: Make changes later

You have two ways in, and both edit the same draft.

  • From your agent: ask, for example, “In my PageHub site [name], add a reviews section under Services.” Check the screenshot, then say “publish.”
  • From the editor: open the site, click Design, make your changes, and publish.

If a change goes wrong, the site's History page lets you restore an earlier version of the whole site or one page. The free plan keeps 15 versions.

Other ways to ship from your MCP client

PageHub is one way to publish a website from your MCP client. You can also have your agent write the site as code in your project, then deploy the code yourself. That's how most people deploy a website from your MCP client today.

  • Vercel or Netlify: connect your Git repository, and each push builds and deploys the site. You add your domain in their dashboard and set DNS at your registrar. A good fit when the site is really an app with its own code.
  • GitHub Pages: serves static files straight from a repository. A good fit for simple static sites and project docs.

The tradeoff: you own every line and can build anything. You also handle Git, builds, and DNS yourself, and every text change is a code change. Someone who doesn't use your MCP client can't easily update it. PageHub fits when you want a site that people can edit visually. It isn't the right tool for custom app logic.

Troubleshooting

PageHub doesn't connect

  • VS Code: run MCP: List Servers, pick PageHub, and choose Start Server, or Show Output to see why it failed. Not listed? Check the file uses servers, not mcpServers. A missing comma or bracket breaks the whole file.
  • Gemini CLI: type /mcp. If it asks you to sign in, run /mcp auth PageHub. If it says DISCONNECTED, run gemini mcp list and check the URL. Not listed? You probably added it without --scope user in another folder.
  • Antigravity: type /mcp for each server's status and log. Check the file is valid JSON at ~/.gemini/config/mcp_config.json and uses serverUrl, then reload from the MCP Manager.

Sign-in doesn't finish

In the browser window, sign in with the email link or the 6-digit code, then click Approve. If the code has expired, request a new one. Gemini CLI and Antigravity sign you in through a browser on the same computer, so that won't work over SSH or in a remote container.

If the browser sign-in still won't complete, use an API key instead. Create one at pagehub.dev/dashboard/api-key and use your client's config below.

VS Code asks for the key once and stores it, so it never sits in the file:

mcp.json (API key)
{
  "inputs": [
    {
      "type": "promptString",
      "id": "pagehub-key",
      "description": "PageHub API key",
      "password": true
    }
  ],
  "servers": {
    "PageHub": {
      "type": "http",
      "url": "https://pagehub.dev/api/mcp",
      "headers": {
        "Authorization": "Bearer ${input:pagehub-key}"
      }
    }
  }
}

Gemini CLI reads it from PAGEHUB_API_KEY in your shell profile. Change PageHub's entry in ~/.gemini/settings.json to:

settings.json (API key)
{
  "mcpServers": {
    "PageHub": {
      "httpUrl": "https://pagehub.dev/api/mcp",
      "headers": {
        "Authorization": "Bearer $PAGEHUB_API_KEY"
      }
    }
  }
}

Antigravity takes the key in the entry, in place of ph_your_key_here:

mcp_config.json (API key)
{
  "mcpServers": {
    "PageHub": {
      "serverUrl": "https://pagehub.dev/api/mcp",
      "headers": {
        "Authorization": "Bearer •••••••••••••••••• "
      }
    }
  }
}

The agent doesn't use the PageHub tools

  • VS Code: check the chat is in Agent mode, then click Configure Tools and make sure PageHub's tools are on.
  • Gemini CLI: servers load when it starts. If you added PageHub while it was running, quit and start it again.
  • Antigravity: if you edited the file while it was running, reload from the MCP Manager.

If you see a rate-limit error, wait a minute. The server allows 120 requests per minute for each sign-in.

The domain isn't working

  • Compare each record at your registrar with what your agent gave you, character for character. A wrong value fails the same way slow DNS does.
  • The bare domain needs an A record. It can't be a CNAME.
  • If your agent gave you a TXT record, add it as well as the others.
  • DNS changes can take a while to spread. Ask your agent to check the domain again later.
  • If the domain is already attached to another Vercel project, remove it there first, then ask your agent to try again.
  • Each site takes 1 domain, plus its www version.

Want the full tool list and other clients? See the MCP server docs or compare plans.

Frequently Asked Questions

Short answers about building and hosting a site from an MCP client.

No. They write and run code, so the site has to be hosted somewhere else. In this guide PageHub hosts it, with SSL included. If you have your agent write the site as code instead, you deploy that code to a host such as Vercel, Netlify, or GitHub Pages.

Build your first site from your MCP client

Make a free account, add PageHub to your MCP client, and ask for the site.

Visual Studio Code and GitHub Copilot are trademarks of Microsoft Corporation and GitHub, Inc. Gemini and Google Antigravity are trademarks of Google LLC. PageHub isn't affiliated with or endorsed by them.