Over the past few days, I’ve been diving deep into the Model Context Protocol (MCP). It’s truly an evolution in how we use "tool calling" as we already had, significantly enhancing the capabilities of AI agents. With MCP, it's now possible to connect to basically any system or external data source, unlocking a whole new level of flexibility and power.
I started experimenting with some existing MCP servers, but I really wanted to build my own to understand how it works in practice. That’s when I had the idea to revisit a super cool project I followed years ago: BrasilAPI, created by Filipe Deschamps.
BrasilAPI is basically Brazil, but in API form. It provides a variety of endpoints for querying data like postal codes (CEP), banks, books/ISBNs, tax rates, brokers, companies, holidays and much more.
So, I decided to create an MCP server that integrates with BrasilAPI. This allows LLMs and AI agents to access up-to-date Brazilian data effortlessly. Some models, depending on where they’re running, don’t have internet access to fetch this kind of information. Others might attempt to answer but often hallucinate and return inaccurate data.
With this MCP server, any LLM can connect and retrieve reliable brazilian data in real time!
It’s still a work in progress, there are still some peding tools to be created, but it’s already available on GitHub for contributions and listed on Smithery and Pulse (hubs for MCP servers) ready to be used in applications!