AG2A: AutoGen Agent-to-Agent

A framework for creating AG2 (formerly AutoGen) agents that communicate using the A2A Protocol and leverage tools via MCP.

Overview

AG2A makes it simple to:

1. Define agents using JSON specifications
2. Connect AG2 agents with MCP-powered tools
3. Expose agents via the standardized A2A protocol
4. Build multi-agent systems that communicate seamlessly

Installation

# Install from PyPI
pip install ag2a

# Install with dev dependencies
pip install ag2a[dev]

Quick Start

Define an Agent Specification

Create a JSON file with your agent's specification:

{
  "name": "MyAssistant",
  "system_message": "You are a helpful assistant that answers questions concisely.",
  "model": "gpt-4o",
  "temperature": 0.7
}

Run the Agent

Start the agent server:

# Export your API key
export OPENAI_API_KEY=your-api-key-here

# Run the agent with your specification
python -m ag2a --spec agent_spec.json --port 8080

Connect to the Agent

The agent now exposes an A2A protocol endpoint at http://localhost:8080. You can connect to it using any A2A-compatible client.

Adding MCP Tools

To use MCP tools, add an mcp_command to your agent spec:

{
  "name": "YoutubeAssistant",
  "system_message": "You can help with analyzing YouTube videos.",
  "model": "gpt-4o",
  "mcp_command": "mcp-youtube"
}

Using Structured Response Models

For agents that need structured output, you can specify a Pydantic response model:

python -m ag2a --spec agent_spec.json --response-model my_module.ResponseModel

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

This project is licensed under the MIT License - see the LICENSE file for details.