I have been building Filament Studio as a dynamic data model manager for Filament v5 and Laravel 12.

The original value prop was runtime-defined collections and fields on top of an EAV model, so you can build flexible admin/data systems without creating a new migration for every content type.

That part worked.

But while working on the project, I kept coming back to a different question:

If this data layer is dynamic, how should an AI agent interact with it?

My answer in v1.3.0 was to add the MCP foundation instead of hacking together one-off AI endpoints.

So this release makes Filament Studio usable as an AI-facing layer, not just an admin-facing one.

What I added in v1.3.0

• MCP server foundation under src/Mcp/
• HTTP/SSE transport mounted at /ai/studio
• stdio transport via php artisan mcp:start studio
• Reuse of StudioApiKey for MCP auth
• New management scope namespace for MCP operations
• Capability discovery resources like:
  • studio://info
  • studio://field-types
  • studio://panel-types
  • studio://operators
• 12 schema-management MCP tools for collections and fields
• Confirm-token flow for destructive operations
• Canonical serializer + exception handling for tool responses
• Tenant-aware behavior aligned with the rest of Studio

Why I think this matters

A lot of AI integration work feels shallow to me.

Either there is a chat UI bolted on top, or there is an agent wired straight into places it probably should not be touching.

For a system like Filament Studio, neither approach feels right.

If the schema is dynamic, the agent needs a real contract:

• discover what exists
• inspect schema safely
• mutate collections and fields through structured tools
• handle destructive actions with explicit confirmation tokens
• work through auth scopes instead of bypassing application boundaries

That is the part I wanted to build first.

What changed for me conceptually

Before this release, Filament Studio was mainly a human-operated system.

After this release, it is starting to become an agent-addressable one.

That is a much more interesting direction than sprinkling AI on top of the UI.

I want the data model, schema rules, and operational boundaries of a Laravel app to be available to agents in a structured way.

v1.3.0 is my first serious pass at that.

What this release is not

This is not the final AI story for the project.

It is the foundation.

I started with schema management because if that layer is weak, everything above it becomes fragile.

So I would rather get the contract right first, then expand the workflows on top of it.

If you are building with Filament and thinking about AI agents beyond demo-level features, this is the direction I am building toward.

Follow the project if you want to see where the MCP roadmap goes next.

Feedback I would like
Filament Studio v1.3.0 is available now.

GitHub: https://github.com/flexpik/filament-studio
Packagist: https://packagist.org/packages/flexpik/filament-studio