A2A

You run an MCP server that acts as an A2A Client for the A2A Protocol, enabling you to connect to remote A2A Agents, send messages, manage conversations, and view artifacts that would otherwise overload context. This guide shows practical steps to use, install, and configure the server so you can start integrating agents into your workflow.

How to use

You interact with the MCP server by connecting through an MCP client. Start the server instance using the provided runtime command, then configure your client to point at the server. Once connected, you can discover available agents, send messages to them, continue conversations, and inspect artifacts that help you reason about long or complex results.

How to install

Prerequisites: you need a compatible runtime (see the runtime setup below). Install the runtime if you haven’t already, then proceed to install and run the MCP server.

# Prerequisite: install the runtime (example shown for uvx-based runtime)
# Follow the runtime’s official installation steps if you don’t have it yet

# Install and run the MCP server using the provided package
# The command shown below is the runtime command and the MCP package name
uvx

Additional configuration and runtime details

Configure your client to load the MCP server by pointing to the Agent Card URL(s) that define the agents you want to connect with. You can specify one or more Agent Cards, each potentially including custom HTTP headers for authentication or other configuration.

{
  "mcpServers": {
    "a2a": {
      "command": "uvx",
      "args": ["a2anet-mcp"],
      "env": {
        "A2A_AGENT_CARDS": "[{\"url\": \"https://example.com/.well-known/agent-card.json\"}]"
      }
    }
  }
}

Agent discovery, messaging, and artifacts

Use the available tools to discover agents, send messages, and view artifacts. This server stores conversations in JSON format for easy inspection and future reference.

Data storage

Agent conversations are stored in JSON format at standard locations depending on your OS. This makes it easy to back up or inspect past interactions.

Development and local testing

For local development, you can run the MCP server in a Claude Desktop workflow. Use the provided runtime command and environment variable to load your Agent Card(s) during development.

Available tools

list_available_agents

Discover available agents and their capabilities.

send_message_to_agent

Send a message to a selected agent to continue a conversation.

view_text_artifact

View text content from an artifact with optional line range.

view_data_artifact

View structured data from an artifact with optional filtering.