An API that AI assistants can use: Laravel, Android and an MCP server

A multi-tenant inventory product built as three parts: a Laravel API, a Kotlin Android app, and an MCP server so an assistant can answer questions about the stock without touching the database.

The situation

A household needs to know what is in stock, where it is and when it runs out. The phone is the natural device, and the question is often spoken or typed in plain language. I built it for my own family, and it is the pattern I sell: one API, several faces.

The approach

• Laravel API with authentication, roles and a Filament admin panel; multi-tenant from the first migration.
• Kotlin and Jetpack Compose app for the phone: barcode scanning, photos, works with a bad connection and syncs later.
• An MCP server (Model Context Protocol) in front of the API: named, permissioned tools such as "what is low", "where is X", so Claude or another assistant answers from the real data, read-only, every call logged.
• Deployed with Docker on a server in Amsterdam, backups and monitoring from day one.

The result

• One codebase answers three clients: app, admin, assistant. A new face is a thin layer, not a rewrite.
• The assistant never sees the database; it sees tools with the user's own rights.
• The same gateway pattern applies to any product that already has an API.

Technology

Laravel, Filament, MySQL, Kotlin, Jetpack Compose, TypeScript MCP server, Docker, DigitalOcean.

More case studies: Phalcon to Laravel, module by module · A small model trained for one job: a wake word · A one-person company run by agents ·