MCP plus business logic is a super power

MCP is shiny and new. Formulize is a 20 year old open source database. But adding MCP to this old school PHP code was a breeze. Let me show you how we did it, crafting tools that AI can understand, and exposing the DB semantics. Giving AI access to the business logic layer was the true super power.

Formulize is a data management system where you configure the business logic, not the database.

You create forms and specify the questions in the forms (which get translated by Formulize into tables and fields). You define the permissions for which users can perform what actions with the form and the data (which get evaluated at runtime before users can perform any actions). You create different screens for different purposes, so users have an interface to work with the forms that is appropriate for your business logic workflow.

Formulize is a mature open source project, and we wanted to update it with AI. Nothing like keeping up with the Jonses.

What we found is that by being 100% configuration based, Formulize is ideal for working with AI. Besides reading the data that users have entered into forms, AI reads the entire configuration and business logic of your application, which means it immediately understands what your application is for, and how it works, and how the pieces are connected together.

And when you modify your application, no need to rewrite any MCP tools or integration, because the AI just reads your configuration changes/additions and it’s ready to go.

To make this possible, we had to create an MCP server and build that into Formulize. We actually used AI (Claude) to write the MCP Server (which is in two parts, a local proxy server, and a remote server built into Formulize, which runs on a LAMP stack).

Then we had to refine it, including revising tools so the AI understood how to use them, and restructuring some of the Formulize configuration so it was more transparent to the AI. But as soon as we did that, we didn’t have to tell the AI how to use all this info, it was amazingly good at taking the ball and running with it, because everything in the system is described in business logic terms, nothing low level.

We want to show you how we created the MCP server, integrating it into this old school LAMP stack application, the pitfalls we ran into, and the lessons learned, from architecture constraints to error handling.

The end result is a 100% free and open source, AI-enabled data management platform, that anyone can use to create any kind of data management application.

Or, now that any MCP-compatible AI assistant can work with Formulize, a platform that AI can use to create any kind of data management application on your behalf. I, for one, welcome our new LLM overlords.

The description pretty much sums it up. Myself and others (including a Montreal-based developer) have been working on the Formulize project for 20+ years. We’ve seen a lot and Formulize has evolved as the web has evolved, and seeing how MCP has the potential to really change everything about how people use systems, is very exciting. We’re really keen to share this with anyone and everyone, and it’s all open source, so have at it folks!