Brazilian Sanctions MCP Server

An MCP (Model Context Protocol) server that provides access to Brazilian sanctions databases through the Porta da Transparência API.

Components

• mcp_server.py - MCP server exposing Brazilian sanctions search tools
• main.py - AI chatbot using LangChain with Ollama for natural language queries
• BrazillianSanctions.json - MCP configuration file

Features

• Search across 5 Brazilian sanctions registers
• Extract specific fields from API responses (names, sanctioning authorities, dates, locations)
• Check if sanctions are still active based on end dates
• Get comprehensive overview with maximum information from all registers
• A simple Natural language interface via AI chatbot (only for tests)
• Search by CNPJ, CPF, or company name
• Detailed sanction information including reasons, dates, and agencies
• Secure API key authentication
• Async-compatible tool execution

Sanctions Registers Supported

Register	Purpose	Search By
CNEP	National Register of Punishments	CNPJ, CPF
CEPIM	Register of Sanctioned Public Employees	CNPJ, CPF
CEIS	Register of Ineligible Entities	CNPJ, CPF
CEAF	Ineligibility Register of Public Servants	CPF
Leniency Agreements	Leniency Agreement Records	Name, CNPJ

Important: Name search is only available for Leniency Agreements. The other registers (CNEP, CEIS, CEPIM, CEAF) only support CNPJ/CPF searches due to API limitations.

API Key Configuration

An API key is REQUIRED to use this system.

Get your API key:

1. Visit: https://www.portaldatransparencia.gov.br/ (Brazilian CPF is required)
2. Login or create an account
3. Go to "Minha Conta" → "API Keys" or "Desenvolvedor"
4. Generate a new API key
5. Copy the key (format: abcd1234efgh5678ijkl9012)

Setup:

Create .env file in project root:

PORTAL_TRANSPARENCIA_API_KEY=your_key_here

SECURITY: Never commit your API key to version control!

Usage

Running the MCP Server

The MCP server can be run standalone or used by MCP clients:

python mcp_server.py

OR

To use in Claude/Cursor use

\BrazillianSanctions.json

Running the Chatbot (The chatbot it's just for test, make sure to setup the MCP server into a real AML Agentic AI enviroment to proper use)

The chatbot provides a natural language interface to query sanctions databases:

python main.py

Example queries:

• "Need to know if 30.477.220/0001-71 is sanctioned by Brazil and if yes why?"
• "Search for Petrobras in leniency agreements"
• "Check if company with CNPJ 11222333000181 has any sanctions"

MCP Server Tools

The MCP server exposes the following tools:

Search Tools

• search_cnep - Search CNEP (National Register of Punishments) by CNPJ, CPF, agency, or date range
• search_cepim - Search CEPIM (Register of Sanctioned Public Employees) by CNPJ, CPF, or superior agency
• search_ceis - Search CEIS (Register of Ineligible Entities) by CNPJ, CPF, agency, or date range
• search_ceaf - Search CEAF (Ineligibility Register of Public Servants) by CPF, agency, or date range
• search_acordos_leniencia - Search Leniency Agreements by name, CNPJ, status, or date range
• search_all_registers - Search all registers for a given CNPJ or CPF

Analysis Tools

• check_sanction_status - Check if an entity is still sanctioned across all registers based on end dates
• get_sanctioning_authorities - Get information about who sanctioned the entity (sanctioning agencies)
• get_location_info - Get location/state information (UF, agency location) for sanctioned entities
• get_comprehensive_overview - Get comprehensive overview with maximum information from all registers including names, authorities, dates, location, and status

Utility Tools

• get_sanction_details - Get detailed information about a specific sanction register type
• check_entity_sanctions - Complete sanctions check for an entity across all registers
• batch_search - Search multiple documents (CNPJs or CPFs) in one operation
• generate_sanctions_report - Generate a formatted text report from multiple search results

Chatbot Demo

I'm testing with a local chatbot based in Ollama but the MCP as you know can be plugged to any Agentic AI:

• LLM: Ollama with qwen3:8b model
• Agent: LangChain create_agent with tool calling
• Tools: Custom LangChain tools wrapping the sanctions API

Architecture

User Query (Natural Language)
    ↓
LangChain Agent (Ollama)
    ↓
Tool Selection & Execution
    ↓
MCP Tools / Direct API Calls
    ↓
Brazilian Sanctions API
    ↓
Detailed Sanction Information
    ↓
Natural Language Response

Document Formatting

The system automatically cleans document numbers(Brazilian CPNJ or CPF). Both formatted and unformatted documents work:

# These are equivalent:
"12.345.678/0001-00"
"123456780001000"
"123.456.789-66"
"12345678966"

Error Handling

The system includes comprehensive error handling:

• API connection errors
• Invalid document formats
• Missing API keys
• Rate limiting
• Malformed responses

Troubleshooting

Issue	Solution
API key not found	Set PORTAL_TRANSPARENCIA_API_KEY in .env file
Ollama not running	Start Ollama service: ollama serve
Model not found	Pull model: ollama pull qwen3:8b
Connection timeout	Check internet connection and API status
No records found	Verify CNPJ/CPF format and check if entity exists

References

• API Documentation: https://api.portaldatransparencia.gov.br/swagger-ui/index.html
• Porta da Transparência: https://www.portaldatransparencia.gov.br/
• MCP Protocol: https://modelcontextprotocol.io/
• LangChain: https://python.langchain.com/
• Ollama: https://ollama.ai/

Last Updated: June 2026 API Base URL: https://api.portaldatransparencia.gov.br/api-de-dados