Brazil implementation

Repo: okfn/mcp-dados-brasil · Source: portaldatransparencia.gov.br · Language: Portuguese

MCP tools over Brasil's emendas parlamentares (parliamentary amendments), published by the Office of the Comptroller General (CGU) on the Portal da Transparência. All tools are Python tools; the plugin declares no YAML datasets.

Highlights

The plugin exposes tools to query amendments by locality, author, government function and subfunction, budget action, and amendment type, plus rankings of top recipients (favorecidos) and top authors, and a glossary with the official definitions from the portal's data dictionary (so the model never confuses empenhado, liquidado and pago values).

How it works

The package ships the amendments dataset (CSV files from the Portal da Transparência download service) and a load-emendas-db script that loads them into a local SQLite database. Each tool runs a SQL query over that database with pandas and returns tables and charts through the standard tool results contract.

Installing it

It is a pip-installable Python package, registered through the mcp_server entry point. Install it into the MCP server's environment, run load-emendas-db once to build the local database, and restart the server. Descriptions, parameters and sample questions are written in Portuguese, matching the audience.

From the field

This catalog backed a pilot with Brasil's Office of the Comptroller General, focused on parliamentary amendments, one of the most requested datasets on the portal. The goal was to test whether citizens could ask in natural language and get answers traceable back to official data. The findings are written up in the Field Guide to Connecting AI to Public Information.