Skip to content
← All notes

· MCP · AI · Open data · Open Source

Itzamná: Mexico's public data, ready for AI

After mcp-banxico I wanted to go beyond a single server. Itzamná is the project that will bring Mexico's official sources together as open-source MCP servers.

Itzamná logo: a stepped fret in purple and blue

When I published mcp-banxico I noticed something: the hard part wasn’t connecting Claude to the Bank of Mexico, it was making every answer say where the figure came from and how current it is. That’s what turns an AI assistant into something you can trust for real decisions.

And Banxico is just one source. Mexico has many more: INEGI, the SAT tax authority, fuel prices, government procurement. Ask any model about them today and it will answer confidently… and often with a stale or made-up number.

The idea

Instead of a single server, I want to build a family of MCP servers, one for each major public data source in Mexico, all following the same rules:

  • No figure without a source. Every answer carries the official series or record and its cut-off date.
  • Open source and free. MIT license, PyPI packages and public endpoints with no sign-up.
  • Made in Mexico, in Spanish, for local use cases: invoices, CETES, inflation, the businesses in your neighborhood.

The first one already exists: mcp-banxico, with over a thousand monthly downloads on PyPI. Next up are mcp-inegi (economic indicators and the directory of 5+ million businesses), mcp-sat (to check whether an invoice is valid and whether the issuer is on the 69-B list) and mcp-gasolina (prices by gas station).

Why Itzamná

Itzamná was the Maya god of wisdom. Tradition says he invented writing and the calendar: the way to keep a record of what is known and when it happened. That’s exactly what I want these servers to do for AI.

The logo is a stepped fret, the xicalcoliuhqui, found on temples all over Mexico.

Where it’s going

  1. MCP servers for the country’s main public data sources.
  2. The Itzamná agent: an assistant that works with any model (Claude, GPT, Gemini or a local one with Ollama) and always checks the official source before answering anything about Mexico.
  3. Later on, an open model fine-tuned to use these tools.

I’m not going to train a model from scratch: that costs millions. The value isn’t in the model, it’s in clean data and in tools any model can use.

The project lives at github.com/itzamna-mx and has its own page here. If you work with Mexican data and there’s a source you’d like to see connected to AI, get in touch.