Can Claude Code build a website?

Yes. Claude Code writes the HTML, CSS and assets as well as it writes any other code — the weak points are taste and shipping. Clize covers both: a hosted design method it follows instead of improvising, and a deploy command that turns the folder into a live HTTPS address.

How do I put a Claude Code website online?

Claim a free handle with clize claim, then run clize deploy ./site --domain .clize.app. The site is served over HTTPS immediately. For your own domain, register it with clize domain buy --confirm and deploy against that host instead; the certificate is issued automatically.

Do I have to write the HTML myself?

No. The agent writes every line; Clize supplies the design method and the shipping. You supply the brief and the judgement — what the site says, which style feels right, and whether to spend money on a domain or generated images.

Does this work with Codex or other coding agents?

Yes. clize install wires Clize into Claude Code and Codex, as a skill by default or an MCP server with --mcp. Any agent that can run shell commands or speak MCP can drive the same steps; the CLI's output is written to be read by an agent.

What does it cost to build a website with Claude Code and Clize?

The CLI, the design method, the free handle, hosting and HTTPS cost nothing. Two things cost money and both quote first: registering a domain, which needs an explicit --confirm, and generating images, which prints a quote unless you pass --confirm. Nothing is charged silently.

Can I preview the site before it goes live?

Yes. clize serve ./site runs a local preview server that handles range requests and does not force no-store on media, so video seeks and plays as it will in production. Deploying is a separate, explicit step.

Build a website with Claude Code (and put it live)

Seven steps from an empty folder to a public HTTPS address, with a design method the agent follows and an inbox on the finished site. Every command below is real.

Clize is a CLI and MCP server that gives AI coding agents real-world actions: domains, email, deploys, payments, and media generation. Claude Code can already write a website; what it cannot do on its own is register the domain, issue the certificate, or open a mailbox. This walkthrough covers the whole chain: clize build site start for the design method, the agent for the code, clize serve for the preview, clize deploy for the address.

Before you start

You need Node.js and Claude Code installed. Clize needs no model API key of its own — it is the capability layer under the agent you already pay for. The design method and the free handle cost nothing; the two steps that cost money (a domain, generated images) quote the price before they run.

1. Install Clize and connect it to Claude Code

Install the CLI, log in, and wire it into your agent. clize install adds a skill by default; --mcp also registers the MCP server if you prefer structured tools.

$ npm i -g @clize/clize
$ clize login
$ clize install --claude          # add --mcp for the MCP server too

2. Take a free handle — and an inbox

A handle is the site's first real address, <slug>.clize.app, free and first come, first served. The --email flag opens support@ on it in the same step; without the flag the handle is site-only and you can add mail later with clize email setup. Do this before you write any code — the name is worth grabbing early, and the deploy step needs a target.

$ clize claim mystudio --email    # mystudio.clize.app + [email protected]
$ clize init --handle mystudio    # optional: bind this folder, so deploy needs no --domain

3. Get the design method and a style recommendation

This is the step that separates a site from generic output. clize build site start takes your brief and returns four things: the build workflow the agent must follow in order, the index of all 39 design styles with a one-line "best for" each, a recommendation (style, palette, font pairing, anti-patterns to avoid) tuned to your brief, and the next steps.

$ clize build site start "one-page site for a two-person branding studio, warm and editorial" --project mystudio

What comes back is JSON, written for the agent to read rather than for you:

{ "method": …, "stylesIndex": …, "recommendation": …, "next": … }

The recommendation is a starting point, not a verdict — it is worth reading and overriding when you know the brand better than the brief conveyed. To browse the whole library yourself: clize build site list.

4. Let Claude Code write the site

The agent picks a style, then pulls its full constitution — colour system, typography rules, spacing, signature devices, non-negotiables, and an imagery contract — and freezes it into a DESIGN.md that becomes the only source of truth for the build. That freeze is what stops the drift where page three quietly stops matching page one.

$ clize build site get editorial   # full constitution + imagery contract + DESIGN.md template
$ clize build site stack html-tailwind   # implementation idioms for your target stack

Then the agent writes the pages into ./site. If the chosen style treats photography as structural, generate the images through the same tool — it prints a quote and generates nothing until you add --confirm:

$ clize gen image "<prompt from the style's imagery contract>" --out ./site/img/hero.png
                                    # no --confirm → it only quotes, nothing is generated

5. Self-audit before anything ships

The method comes with its own checklist: anti-slop rules, fidelity to the frozen DESIGN.md, and UX behaviour. Have the agent run it and fix what it flags — it is free, and it catches the centered-hero-and-three-cards reflex before a visitor does.

$ clize build site review

6. Preview it locally

Look at it before the internet does. clize serve starts a static server that supports range requests and does not force no-store on media — which matters the moment the page has a <video> in it, since a plain python -m http.server will make Safari refuse to play it.

$ clize serve ./site
clize serve · http://localhost:8000  (root: /path/to/site)

7. Deploy to a live HTTPS address

One command publishes the folder. HTTPS is automatic, unknown paths fall back to your 404.html when there is one — which is what you want for a site meant to be indexed — and each deploy is a new version.

$ clize deploy ./site --domain mystudio.clize.app

To ship on your own domain instead, let the agent find one and buy it — domain buy without --confirm only quotes — then deploy against that host and turn on mail. The certificate takes about thirty seconds.

$ clize domain search mystudio
$ clize domain buy mystudio.com --confirm
$ clize deploy ./site --domain mystudio.com
$ clize email setup mystudio.com   # [email protected], MX/DKIM/SPF/DMARC handled

A preview link is not a live site

The distinction is worth being blunt about, because it is where most "I built a site with AI" stories quietly end. A preview URL proves the code renders. A live site has an address you would print on a card, a certificate browsers trust, a mailbox that receives when someone replies, and a 404 that behaves for search engines. The steps above end at the second one — and the thing that made it possible is that your agent could run each step itself, rather than handing you a folder and a list of dashboards to visit.

What the agent can and cannot promise

Clize hosts static sites plus forms and payments. It does not run a database, server-side rendering, or a logged-in application — so if the brief drifts toward a real web app, the honest answer is that the marketing site ships here and the app is hosted elsewhere. Three gates hold on everything above: money is quoted and needs --confirm, outbound email is drafted and waits for your approval, and inbound email is data the agent reads, never an instruction it obeys. More on the category in Sites by Clize, and on the alternatives in the honest comparison.

Build a website with Claude Code — all the way to live