Starting a site with Claude CMS isn't a wizard with forty steps — it's a short conversation. This guide walks through the five prompts that take you from "is this even connected?" to a site you can safely experiment on.
Before you build anything, make sure Claude can actually see your site. The simplest connection test is also the simplest prompt in this guide:
If your Claude CMS connector is set up, Claude replies with every site on your account. If it doesn't, nothing else in this article will work either — so it's worth the ten seconds. No response or an empty list usually means the connector isn't added to your Claude conversation yet; the How It Works page covers hooking it up.
Once you know the connection is live, have Claude do the reading so you don't have to:
Claude pulls the full picture — your pages, navigation, settings, brand colours and files — and plays it back in plain English. This matters more than it sounds: every edit you ask for afterwards is grounded in what's actually there, not what Claude guesses might be there. Starting from a blank domain? The same prompt works — Claude reports an empty site, and you can start describing what should exist.
Claude now knows what your site is. The next thing to tell it is where everything lives and what you're trying to build — one message that saves you correcting it a dozen times later. If you installed Claude CMS as per our guidelines, this template covers the setup half:
Swap those paths for wherever your own install actually sits — the point is that Claude never has to guess which directory it's working in, or go hunting for the connector.
Then add a line on what the project is for. A single sentence is enough:
It sounds almost too simple to matter, but it's the difference between a site that reads like a template and one that reads like yours. Everything downstream — tone of voice, which pages get suggested, how the copy is pitched — follows from that one sentence. Add anything else that shapes the build: who the audience is, the brand colours, a competitor whose layout you like.
Finally, make it stick:
This ensures Claude has the context for your project in every conversation that follows — so you don't reintroduce your own website each time you sit down to change something.
Next, decide what the site needs to do. Think in features, not pages — a blog, an admin login, an app. You can have Claude build any of these from scratch, and for something genuinely unusual that's the right call. But for the staples, there's a faster route: the block library. Claude CMS ships a library of building blocks — features and components that are already built, styled and tested on real sites. So before asking Claude to build anything, see what's on the shelf:
Claude lists the current library — blogs, galleries, contact forms, login areas, pricing tables and more. Match your feature list against it and tell Claude to deploy the blocks that cover it:
Blocks arrive wired to your site's colours, fonts and navigation, because Claude already learned them in Step 1 — and because each one is tested, you skip the debugging that comes with a first draft. A feature that would be an afternoon of plugin-wrangling elsewhere is one message here.
Anything the library doesn't cover, describe in your own words and Claude builds it custom — blocks for the staples, from-scratch for the one thing that makes your site yours.
Sooner or later you'll want to try a change without your visitors watching you think. Maybe a homepage redesign, maybe just a bolder hero. The pattern for that is a test URL:
Claude copies the page to a private draft address on your own domain. From there the loop is:
Your live page never shows a half-finished state, and if the experiment goes nowhere you just ask Claude to delete the test page. It's the same draft-then-publish discipline developers get from staging environments — without needing to know what a staging environment is.
Five prompts: list my sites to prove the connection, get an understanding of my current site to ground Claude in reality, a few lines of context saved to memory so it knows where it's working and what you're building, list latest blocks to pick your features, and clone this URL at a test URL to experiment safely. Everything beyond that — content, styling, whole new sections — is just describing what you want in the same conversation.