MULTI-AGENT TRADING SYSTEMS APPLIED IN THE BRAZILIAN FINANCE MARKET

This project presents the design, implementation, and evaluation of a Multi-Agent Trading System using a structured Model Context Protocol (MCP) for agent communication. The system integrates reactive agents, planning agents, committee agents, guardrail agents, and an execution agent, orchestrated to process market data, generate trading signals, evaluate strategies, enforce risk controls, and execute trades. Historical financial data from Yahoo Finance was used to simulate real-market scenarios, and a simple backtesting framework was applied to evaluate performance. The agents communicate via MCP messages, enabling a modular and extensible architecture. Over a predefined number of cycles, the system produced trade signals and executed orders, demonstrating the feasibility of a multi-agent approach for algorithmic trading. The test highlighted areas for improvement, including signal generation accuracy, risk-adjusted strategy evaluation, and performance optimization. The project also explored Docker-based deployment for environment reproducibility and scalability, providing a framework for multi-asset and parallelized trading simulations.