How AutoDev Leverages Google’s A2A Protocol for Cross‑Agent Collaboration

Overview of A2A Protocol

The Agent‑to‑Agent (A2A) protocol, introduced by Google, enables direct communication between autonomous AI agents. It allows a task to be packaged as a message and delegated to another agent for execution, facilitating cross‑agent and cross‑system collaboration.

Simple A2A Example

Message message = A2A.toUserMessage("tell me a joke");
jokeClient.sendMessage(message);

This example shows how a textual request is wrapped in a Message object and sent to a remote agent.

Integration with AutoDev and MCP

AutoDev already provides bidirectional Model Context Protocol (MCP) support: it can act as an MCP server for any Agent Tool and as an MCP client to invoke external tools. Adding A2A extends this model, allowing AutoDev Sketch to delegate tasks to an MCP Tool or to a remote AI agent.

MCP : Standardizes communication between coding agents and tools, enabling direct use of external APIs and data sources.

A2A : Standardizes communication between coding agents themselves, simplifying remote AI agent integration and reducing parameter‑related errors.

AutoDev A2A Implementation

Prompt assembly – A2ASketchToolchainProvider converts an AgentCard into an AgentTool, providing a unified tool interface and prompt for the AI.

Tool invocation – A2AInsCommand sends the message to the target AI agent, which then performs the requested action.

Configuring AutoDev for A2A

Add an a2aServers section to AutoDev’s configuration file to specify remote agent endpoints.

{
  "mcpServers": {
    // existing MCP server definitions
  },
  "a2aServers": {
    "remote-example": {
      "url": "http://localhost:9999"
    }
  }
}

After the configuration is in place, reference the new AI agent in a DevInsCommand:

/a2a:remote-example
{
  "agent": "Hello World Agent",
  "message": "Please review this code for potential security vulnerabilities"
}

AutoDev A2A Debugger

Create a *.mcp.json file, switch the UI to A2A mode, and you can view detailed request/response information and send test messages.

Key Benefits

Cross‑agent, cross‑system task delegation without tight coupling.

Open‑source architecture enables community‑driven refactoring and modularization.

Provides a platform for interoperable agents, tools, and protocols.

GitHub repository: https://github.com/unit-mesh/auto-dev