This repository explores multi-agent using A2A protocol using various agentic development frameworks:

• Google ADK
• Microsfot Agent Framework
• A2A JS SDK
• Langchain MCP
• BeeAI Framework
• Agent Stack

Data and frameworks based on Deep Learning A2A course

Project Structure

A2A-Protocol-Sample
├── pythonA2A                         
│   └── src/
│       ├── clients/           # A2A clients using BeeAI Agent Framework           
│       └── servers/           # BeeAI Agent Stack servers wrapping around A2A clients. Run as a standalone chat app via Agent Stack Web UI.
│           └── agentstack/                     
├── pythonAdk                  # A2A clients using Google Adk. Run as a standalone chat app via ADK web UI.
│   └── agent/                 # Policy and research A2A client agents
└── tsServers                  # A2A servers (remote agents)   
│   ├── data/                  # "database"    
│   └── src/
│       ├── a2aServer/         # A2A servers implementation
│       └── mcpServer/         # MCP servers implementation

Setup

• Node.js
• Python >= 3.11
• Agent Stack
• API Keys in tsServers:
  • GOOGLE_GENAI_API_KEY
  • ANTHROPIC_API_KEY
• API Keys in PythonA2A:
  • GEMINI_API_KEY

Run A2A TypeScipt servers

# In tsServers folder
npm install
npm run build
npm run server

Run Full Concierge with AgentStack

1. Run A2A Python clients on Agent Stack

# In pythonA2A folder
uv sync
uv run python src/servers/run_all_agentstack.py

2. Launch Agent Stack Web UI

agentstack ui

Select Healthcare Concierge

Run A2A Clients on ADK

# In pythonAdk
uv sync
uv run adk web