A multi-agent system for intelligent stock analysis and trading decisions, TradingAgents-MCPmode integrates the Model Context Protocol (MCP) to provide comprehensive market analysis, investment advice, and risk management. It benefits technical users and researchers by offering a collaborative framework of specialized AI agents for advanced financial insights.

How It Works

The system employs a sophisticated multi-agent architecture comprising 15 specialized agents organized into Analyst, Researcher, Manager, and Risk Management teams. A core innovation is the parallel execution of six specialized analysts (Company Overview, Market, Sentiment, News, Fundamentals, Shareholders, Product), significantly boosting efficiency. These analysts leverage the MCP tool for real-time data acquisition. Subsequent stages involve intelligent debate mechanisms between bull/bear researchers and risk analysts, configurable debate rounds, and dynamic agent control, all orchestrated to refine investment strategies and manage risk.

Quick Start & Requirements

• Installation: Clone the repository, then install dependencies via pip install -r requirements.txt.
• Environment: Python 3.8+ on Windows, macOS, or Linux.
• Prerequisites: Requires API keys for OpenAI (e.g., OPENAI_API_KEY) and Tushare (configured in mcp_config.json).
• Configuration: Set API keys and workflow parameters in .env and MCP server details in mcp_config.json.
• Web UI: Run streamlit run web_app.py for an interactive experience at http://localhost:8501.
• Demo: A public demo is available at http://47.79.147.241:8501.

Highlighted Details

• Multi-Agent Collaboration: Features 15 specialized agents for comprehensive analysis and decision-making.
• Parallel Analyst Architecture: Six distinct analysts execute concurrently, drastically reducing analysis time.
• MCP Integration: Utilizes the Model Context Protocol for accessing external data sources and real-time information.
• Intelligent Debate: Incorporates configurable debate rounds between bull/bear researchers and risk analysts to refine insights.
• Dynamic Control: A Streamlit web frontend allows real-time enabling/disabling of agents and adjustment of debate parameters.
• Multi-Market Support: Capable of analyzing stocks across US, China (A-shares), and Hong Kong markets.
• Natural Language Interface: Supports queries in natural language, abstracting market and date specifics.

Maintenance & Community

The project welcomes contributions via Issues and Pull Requests. Contact information includes an email (guangxiangdebizi@gmail.com) and links to personal social media profiles (LinkedIn, Facebook, Bilibili).

Licensing & Compatibility

The project is licensed under the MIT License. This permissive license generally allows for commercial use, modification, and distribution, including within closed-source applications.

Limitations & Caveats

The public demo instance requires "civilized use" and warns against attacks or stress testing, indicating it is a shared, potentially unstable resource. The system relies on external API keys (OpenAI, Tushare), which incur costs and require careful management. While theoretical performance gains from parallelization are significant, actual efficiency depends on the execution time distribution of individual agents.