Building a Semantic Layer over Brazil’s Official Economic Data

Description and use

A semantic layer was built over Brazil’s official economic data, unifying disparate APIs while preserving economic meaning and provenance, enabling AI agents to retrieve accurate results.

Problem

Using Brazil’s official economic data programmatically is hard because each source exposes data differently and semantic meaning is lost when normalizing APIs.

Outcome

Open Economics unifies disparate Brazilian economic data sources, preserves semantic meaning and provenance, and enables AI agents to retrieve accurate, machine-readable results.

Workflow

• 1. Define semantic resolution to map user query to economic concept (e.g., "current Selic target rate" → "selic-target").
• 2. Identify available official datasets that represent the concept (e.g., BCB series bcb-sgs:432).
• 3. Retrieve metadata and provenance (institution, dataset ID, unit, frequency).
• 4. Fetch observations via REST or MCP.
• 5. Return structuredContent plus bounded textual representation for AI agents.
• 6. Agents use only Open Economics to answer queries.
• 7. Validate interoperability by testing with Cline agent.

Evidence

The hardest problem was preserving economic meaning.

Source

The remote MCP server currently exposes 19 tools.

Source