Raşit

Posted on Feb 13

I turned my headless CMS into an AI-native CMS. Here’s what happened.

#ai #elmapicms #mcp #webdev

I’ve been building ElmapiCMS for a while. It’s a Laravel + React headless CMS for developers.

But something kept bothering me.

AI tools were exploding everywhere, yet most CMS platforms were just adding a simple “Generate text” button and calling it AI integration.

So I decided to go deeper and turn the CMS itself into something AI-native.

Here’s what changed.

1. From text generation to schema-aware AI

Instead of generating random paragraphs, I made the AI aware of:

Now when you generate content, the output is influenced by the actual structure of your data model.

That small shift made a big difference in quality and relevance.

I added an AI Chat Assistant that can:

You can describe a “blog with title, slug, content and author relation” and it scaffolds it.

This was technically much harder than adding content generation, because it touches schema integrity and validation.

Since the structure is already defined, I added full entry translation per locale.

Instead of translating field by field, you translate the whole entry in one action.

For agencies handling multilingual clients, this is powerful.

This was the most interesting part.

I implemented an MCP server so external AI agents can interact directly with the CMS:

Create collections
Modify schema
Insert content
Query structured data

The idea is to support vibecoding workflows, where AI doesn’t just write frontend code but can shape the backend in a controlled way.

Technically, this is the strongest version I’ve built.

Commercially? Visibility is still hard.

Adding AI features does not magically solve distribution.

What I learned:

I’m still figuring out the go-to-market side.

If you’re building in the CMS, AI tooling, or developer infrastructure space, I’d love to hear from you.