What happens when you build for AI integration… before AI integration was a thing?
As COO and Architect at Finmars, I spent the past month on a personal experiment: Could I build a ready-to-use MCP server for our financial portfolio management platform as if I were just another community developer?
Turns out, yes! And it happened faster than I expected.
But here’s what really caught my attention that architectural decisions we made years ago at Finmars had accidentally set us up perfectly for this moment:
· OpenAPI boundaries everywhere – consistency across the board
· GraphQL as our semantic layer – we speak domain language, not just tech specs
· Open source from day one – AI agents can read actual code instead of guessing
These decisions, made long before MCP existed, accidentally created the perfect foundation for AI integration.
The experiment became a reference implementation:
• Complete MCP server with portfolio management tools
• Investor profiling prototypes (structured context → auditable analysis)
• Working examples of AI-finance integration patterns
To be clear: this isn’t an official Finmars product (yet). It’s architectural R&D that demonstrates what the platform enables. The patterns are real, the code works, and it will be open-sourced as a starting point for the community.
The takeaway? Read more about it here:
