Build a client website with Claude Code
and an MCP-connected CMS.
A coding agent can write files. Connected to a CMS over MCP, it can also create the content model, the pages and the entries behind them. Here is the workflow on Quilium.
In short
You connect Claude Code to Quilium's hosted MCP server, install two process skills, and point the agent at an HTML prototype. It proposes a content model, you correct and approve it, and it writes the YAML, the Liquid templates, the pages and the content. The agent builds, a human publishes.
1. Connect the MCP server
Quilium runs a hosted MCP server that authenticates over OAuth: nothing to install, no token to paste. Add it, then run /mcp to authorise in the browser. The agent inherits your access, with the profile you hold on each site.
claude mcp add --transport http quilium https://mcp.quilium.io/mcp
2. Install the skills, state the rules
Run npx skills add quiliumio/quilium-skills -y once per project. It installs quilium-builder and quilium-webmaster, which teach the process around the CMS; the agent fetches the documentation skills from the MCP server by itself. Then put your ground rules in CLAUDE.md at the project root, where you start Claude Code:
- Before any CMS operation, call get-skills, load every skill relevant to
the task, then get-site-settings for this site.
- Match the prototype's markup and its class names. Do not invent a
section, a component or a piece of copy that is not in it.
- Never publish. I decide what goes live.
3. Ask for a content model, and review it
This step decides everything downstream, so ask for a proposal, not a build. What comes back is a list of collections, content-types and page-types. Correct it in plain words: who owns a value, what will change over time, what belongs to the design.
Read the prototype in proto/ and design the architecture for this site.
Show me a simple bullet list for validation, one line per type: what it is
and where it is used.
Write nothing to the site before I approve it.
4. Let it build, one narrow request at a time
Once you approve, the agent writes the YAML model through the MCP server, then the routes, the pages and the content, while templates are ordinary .liquid files it writes in the project folder. Keep each request narrow: "Now the five elements of the home page. Nothing else." helps keep an agent with momentum from carrying on into pages you have not discussed.
5. Review the drafts, publish yourself
Make verification part of the task instead of trusting "done", then open the back office and edit a block the way your client will. Pages stay in draft until you publish them; collection entries do not, so ask to see the first entry before the agent writes the rest.
When you're finished, run `quilium run`, then curl each page and confirm:
HTTP 200, no unrendered Liquid ({{ or {% in the output), no [object Object],
and every image returning 200. Paste what you checked.
The full walkthrough, with a public starter repository, begins at the AI tutorial.
Run it on your next prototype.
Create an account, connect your agent, and follow the tutorial with its public starter repository. Building is free, you pay when the site goes live.