ADK_MCP_A2A

# Multi-Agent Financial Advisor Management System A sophisticated multi-agent financial advisory system built using Model Context Protocol(MCP), Agent to Agent protocol(A2A) and Google Agent Development Kit(ADK). ## 1. Setup ### Prerequisites - Python 3.9+ - Conda package manager - Poetry dependency manager - Google API key ### Installation Steps 1. **Clone the repository:** ```bash git clone https://github.com/meetrais/ADK_MCP_A2A.git cd ADK_MCP_A2A ``` 2. **Set up environment variables:** - Create a `.env` file inside the `financial_advisor_management/financial_advisor` directory - Add the following entry to the `.env` file: ``` GOOGLE_API_KEY=YourGoogleAPIKeyHere ``` 3. **Navigate to the project directory:** ```bash cd financial_advisor_management ``` 4. **Create and activate conda environment:** ```bash conda create --name ama conda activate ama ``` 5. **Install dependencies:** ```bash poetry install ``` 6. **Run the system:** Open multiple terminals in your IDE and run the following commands in separate terminals: **Terminal 1 - MCP Server:** ```bash poetry run python financial_advisor/simple-mcp-server/mcp_server.py ``` **Terminal 2 - Data Analyst Agent:** ```bash poetry run python -m financial_advisor.sub_agents.data_analyst.agent ``` **Terminal 3 - Execution Analyst Agent:** ```bash poetry run python -m financial_advisor.sub_agents.execution_analyst.agent ``` **Terminal 4 - Risk Analyst Agent:** ```bash poetry run python -m financial_advisor.sub_agents.risk_analyst.agent ``` **Terminal 5 - Trading Analyst Agent:** ```bash poetry run python -m financial_advisor.sub_agents.trading_analyst.agent ``` **Terminal 6 - Web Interface:** ```bash adk web ``` ## 2. Code Overview ### System Architecture The Financial Advisor Management System is built on a multi-agent architecture where specialized agents collaborate through both the MCP protocol and Agent-to-Agent (A2A) HTTP communication to provide comprehensive financial advisory services. <img width="1211" height="653" alt="image" src="https://github.com/user-attachments/assets/9e885142-19bc-4253-8672-3d8b8e9921cc" /> #### Core Components **1. Main Financial Advisor Agent (`financial_advisor/agent.py`)** - **Role**: Central orchestrator coordinating between specialized sub-agents - **Framework**: Built on Google ADK using Gemini 2.5 Flash model - **Communication**: Uses A2A proxy agents to communicate with sub-agents via HTTP - **Architecture**: Implements `A2AAgentProxy` class for seamless integration with external agents - **Port Configuration**: - Data Analyst: `localhost:8080` - Execution Analyst: `localhost:8081` - Trading Analyst: `localhost:8082` - Risk Analyst: `localhost:8083` **2. MCP Server (`financial_advisor/simple-mcp-server/mcp_server.py`)** - **Purpose**: Provides financial analysis tools and mock data services - **Technology**: Flask-based REST API server - **Port**: `localhost:3001` - **Endpoints**: - `/analyze` - Market data analysis for tickers - `/execution-analyze` - Execution strategy analysis - `/trading-analyze` - Trading strategy analysis - `/risk-analyze` - Risk assessment analysis - `/health` - Health check endpoint - **Data**: Generates realistic mock financial data for demonstrations #### Specialized Sub-Agents **1. Data Analyst Agent (`sub_agents/data_analyst/agent.py`)** - **Purpose**: Market data collection, processing, and analysis - **Port**: `8080` - **Integration**: Communicates with MCP server for market data - **Capabilities**: - Stock ticker analysis - Price and volume data processing - Market trend analysis - Trading recommendations - **Communication**: Provides A2A endpoints (`/chat`, `/agent-card`, `/health`) **2. Risk Analyst Agent (`sub_agents/risk_analyst/agent.py`)** - **Purpose**: Comprehensive risk assessment and management - **Port**: `8083` - **Integration**: Uses MCP server for risk calculations - **Capabilities**: - Portfolio risk scoring (1-10 scale) - Value-at-Risk (VaR) calculations - Stress testing scenarios - Diversification analysis - Correlation analysis - Hedging recommendations - **Output**: Detailed risk assessment with volatility metrics and hedging suggestions **3. Trading Analyst Agent (`sub_agents/trading_analyst/agent.py`)** - **Purpose**: Trading strategy development and market analysis - **Port**: `8082` - **Integration**: Leverages MCP server for trading analysis - **Capabilities**: - Trading strategy generation (Buy & Hold, Momentum, Mean Reversion, etc.) - Entry and exit point identification - Risk-reward ratio calculations - Position sizing recommendations - Multi-timeframe analysis (short, medium, long-term) **4. Execution Analyst Agent (`sub_agents/execution_analyst/agent.py`)** - **Purpose**: Trade execution optimization and portfolio management - **Port**: `8081` - **Integration**: Uses MCP server for execution analysis - **Capabilities**: - Order type recommendations (Market, Limit, Stop-Loss, etc.) - Execution timing optimization - Transaction cost analysis - Broker recommendations - Market impact assessment ### Key Features #### Agent-to-Agent Communication (A2A) - **HTTP-based Protocol**: Each agent runs as an independent Flask server - **RESTful Endpoints**: Standardized `/chat`, `/agent-card`, and `/health` endpoints - **Proxy Architecture**: Main coordinator uses `A2AAgentProxy` for seamless communication - **Async Support**: Built on ADK's async event system for non-blocking operations - **Error Handling**: Comprehensive error handling and timeout management #### Financial Analysis Pipeline 1. **Data Collection**: Data Analyst gathers market information via MCP server 2. **Risk Assessment**: Risk Analyst evaluates portfolio risks and exposures 3. **Strategy Development**: Trading Analyst develops investment strategies 4. **Execution Planning**: Execution Analyst optimizes trade implementation 5. **Coordination**: Financial Coordinator orchestrates the entire process #### Structured Financial Advisory Process Based on the system prompts, the workflow follows a methodical approach: 1. **User Introduction**: System presents capabilities and disclaimers 2. **Market Analysis**: User provides ticker symbol for analysis 3. **Strategy Development**: User specifies risk tolerance and investment horizon 4. **Execution Planning**: System develops optimal execution strategy 5. **Risk Evaluation**: Comprehensive risk assessment of the complete plan 6. **Markdown Output**: Results presented in structured markdown format ### Technology Stack - **Framework**: Google ADK (Agent Development Kit) - **LLM Model**: Gemini 2.5 Flash - **Communication**: - A2A Protocol (HTTP/REST) - MCP (Model Context Protocol) - **Language**: Python 3.9+ - **Web Framework**: Flask - **Dependency Management**: Poetry - **Environment Management**: Conda - **Data Format**: JSON for inter-agent communication ### Project Structure ``` financial_advisor_management/ ├── financial_advisor/ │ ├── agent.py # Main coordinator with A2A proxy system │ ├── prompt.py # System prompts and workflow instructions │ ├── simple-mcp-server/ # MCP server implementation │ │ └── mcp_server.py # Flask server with financial analysis tools │ ├── sub_agents/ # Specialized financial agents │ │ ├── data_analyst/ # Market data analysis (port 8080) │ │ │ ├── agent.py # A2A-enabled data analysis agent │ │ │ └── prompt.py # Data analyst instructions │ │ ├── execution_analyst/ # Trade execution (port 8081) │ │ │ ├── agent.py # A2A-enabled execution agent │ │ │ └── prompt.py # Execution analyst instructions │ │ ├── risk_analyst/ # Risk assessment (port 8083) │ │ │ ├── agent.py # A2A-enabled risk analysis agent │ │ │ └── prompt.py # Risk analyst instructions │ │ └── trading_analyst/ # Trading strategies (port 8082) │ │ ├── agent.py # A2A-enabled trading agent │ │ └── prompt.py # Trading analyst instructions │ └── tools/ # Shared utilities (currently empty) ├── pyproject.toml # Poetry configuration with ADK dependencies └── README.md # This documentation ``` ### Agent Communication Flow 1. **User Request**: Client interacts with financial coordinator via ADK web interface 2. **Agent Orchestration**: Coordinator determines which sub-agents to invoke 3. **A2A Communication**: HTTP POST requests sent to sub-agent `/chat` endpoints 4. **MCP Integration**: Sub-agents call MCP server for specialized financial analysis 5. **Response Assembly**: Results aggregated and formatted by coordinator 6. **Client Delivery**: Final analysis presented to user in markdown format ### Dependencies **Core Dependencies:** - `google-cloud-aiplatform[adk,agent-engines]` - ADK framework - `google-genai` - Gemini model integration - `google-adk` - Agent development tools - `flask` - Web framework for A2A communication - `requests` - HTTP client for inter-service communication - `pydantic` - Data validation and parsing - `python-dotenv` - Environment variable management **Development Dependencies:** - `pytest` - Testing framework - `black` - Code formatting - `pytest-asyncio` - Async testing support This architecture provides a robust, scalable, and maintainable foundation for comprehensive financial advisory services through specialized agent collaboration.