Base Dos Dados

A Model Context Protocol (MCP) server that provides AI-optimized access to Base dos Dados, Brazil's largest open data platform.

💡 Create Reports with Real Brazilian Datasets with Claude Desktop

Ask it to replicate a news article:

User:
https://www.gov.br/secom/pt-br/assuntos/noticias/2025/05/no-melhor-abril-do-novo-caged-brasil-gera-257-mil-vagas-com-carteira-assinada

consegue replicar a materia e checar os numeros?

Claude: https://claude.ai/public/artifacts/15d7f5c5-f017-4a96-9380-a93e535001fd

Or a research question

User:
https://claude.ai/share/f8dbbe5b-1c34-4804-9462-1bfb9008558d

Claude: https://claude.ai/share/f8dbbe5b-1c34-4804-9462-1bfb9008558d

✨ Features

• 🔍 Smart Search: Portuguese language support with accent normalization
• 📊 BigQuery Integration: Direct SQL execution and table references
• 🇧🇷 Brazilian Data Focus: IBGE, RAIS, TSE, INEP datasets
• 🤖 AI-Optimized: Single-call comprehensive data retrieval

🚀 Quick Install to Claude Desktop

Just run this line in your terminal

bash -i <(curl -LsSf https://raw.githubusercontent.com/JoaoCarabetta/basedosdados-mcp/refs/heads/main/install.sh)

Add your BigQuery project_id, location and service account.

Restart Claude Desktop and you are good to go!

⚠️ I just tested it in my Mac M-Series

🛠️ Tools

Tool	Description
search_datasets	Search datasets with Portuguese support
get_dataset_overview	Get complete dataset overview with tables
get_table_details	Get table details with columns and SQL samples
execute_bigquery_sql	Execute SQL queries directly
check_bigquery_status	Check BigQuery authentication

🔧 Development

Quick Development Setup

Set up the MCP server for local development with live code reloading:

git clone https://github.com/JoaoCarabetta/basedosdados-mcp
cd basedosdados-mcp
./dev_install.sh

This script will:

• ✅ Install all dependencies in development mode
• ✅ Create a development wrapper script with live reloading
• ✅ Configure Claude Desktop with basedosdadosdev server
• ✅ Set up proper environment variables and paths

Manual Development Setup

If you prefer manual setup:

# Clone and install dependencies
git clone https://github.com/JoaoCarabetta/basedosdados-mcp
cd basedosdados-mcp
uv sync --extra dev

# Run development server
uv run basedosdados-mcp-dev

# Or test directly
./run_dev_server.sh

Development Features

• 🔄 Live Code Reloading: Changes to src/ are reflected immediately
• 🐛 Debug Logging: Enhanced logging for development
• 🧪 Full Test Suite: Comprehensive pytest-based tests
• 🎯 Separate Server: basedosdadosdev server in Claude Desktop
• ⚡ Fast Development: No need to reinstall after code changes

🧪 Testing

Install development dependencies and run the comprehensive test suite:

# Install dev dependencies (includes pytest)
uv sync --extra dev

# Run all encoding tests
pytest tests/ -v

# Run specific test categories
pytest tests/ -k "encoding" -v        # Encoding tests only
pytest tests/ -k "live" -v            # Live server tests only
pytest tests/ -m "not live" -v        # Skip live tests (faster)

# Run with coverage
pytest tests/ --cov=basedosdados_mcp --cov-report=html

Test Structure

The pytest-based test suite includes:

• test_encoding_pytest.py: Core Portuguese character encoding tests

  • UTF-8 encoding/decoding validation
  • JSON serialization integrity
  • MCP protocol compatibility
  • Parametrized tests for individual Portuguese words

• test_live_mcp_pytest.py: Live MCP server endpoint tests

  • Real API response validation
  • Backend API encoding verification
  • Large response handling
  • End-to-end encoding flow

• conftest.py: Shared fixtures and pytest configuration

  • Common test data and utilities
  • Automatic test marking and organization

Encoding Validation

Portuguese characters tested: população, educação, saúde, região, satélites, políticas, públicas, and 30+ more common Brazilian dataset terms.

Corruption patterns detected: é, á, í, ó, ú, ã, ç, ô, ê, à , õ (common UTF-8 corruption)

If you experience encoding issues in Claude Desktop (e.g., seeing satélites instead of satélites), run the tests to verify the issue is in the display layer, not the MCP server itself.

📚 About

Base dos Dados is Brazil's largest open data platform, providing standardized access to Brazilian public datasets through BigQuery.

Data Coverage: Demographics, Economics, Education, Politics, Health, Environment

📄 License

MIT License