Minas Gerais MCP for AI. Automate discovery of state public records.

Minas Gerais (Estado) MCP Server connects your AI client straight into the official Open Data Portal for Minas Gerais. You're not dealing with raw data; you're querying the metadata—the structural blueprint of the state's government records. Forget web scraping or wrestling with complex session management. This server gives your agent direct access to underlying API endpoints, letting it programmatically query, search, and pull organizational details from the entire portal without clicking a single page.

The whole point here is treating the massive transparency database like a structured resource within your workflow. It's about indexing government data for machines, not humans. You run functions that do the heavy lifting, returning clean JSON objects you can pass to subsequent steps in your agent's process.

Searching and Discovery:

If you don’t know what you need, you start broad. To get a full inventory of everything available, run list_packages for an immediate list of every dataset name and its basic metadata in the entire Minas Gerais portal. You can narrow that scope down quickly by running search_packages; just feed it keywords—like 'health' or 'finance'—and you'll get package metadata that matches those specific terms across the whole catalog.

For a more targeted search, use search_resources. If you know you’re looking for something like a 'CSV' file type or data related to 'budget,' this function searches all available resources and files inside the portal.

Understanding Structure:

You need to map out who owns what. Use list_organizations first; it provides a complete list of every department or state body that contributed data. Once you've identified an organization, run get_organization to grab its full details and see precisely which datasets that specific entity manages. To understand the thematic filing system—the official buckets used across the portal—you can call list_groups, which returns a list of all available thematic groups (categories).

If you need more depth on those categories, use get_group to retrieve detailed metadata and associated datasets for any specific group. Similarly, to understand how data is tagged, run list_tags to get every standardized tag used for classification across multiple packages, which helps your agent filter down the search criteria.

Drilling Down into Data Assets:

When you've located a dataset or want deep details on a single file, these tools take over. If you have an exact package ID and need all associated metadata—the complete record for that data set—you execute get_package. That gives you the full picture of the dataset itself. To find specific files (resources) within a known package or across the whole system, use search_resources to locate content types like 'CSV' or 'PDF'.

For an individual file asset, running get_resource retrieves specific metadata about that single resource within any given dataset.

How It Works In Practice:

Your agent runs these tools. It doesn't render a messy webpage; it receives structured JSON responses containing the precise metadata, identifiers, and links you need. You take those IDs—whether they’re package IDs or organization UUIDs—and pass them into other steps in your workflow. This lets your agent build complex data maps, allowing you to trace ownership from a specific file back through its resource container, up to the department that published it.