Simple A2A Agent

A tool to quickly spin up simple Agent-to-Agent (A2A) servers.

Features

• Run A2A servers from CLI parameters or a JSON config file
• Autodetect available port (if not configured)
• Select LLM model and model parameters to use (through VoltAgent)
• Add MCP tools to the agent (through VoltAgent)

WARNING: Experimental

This project was done for educational purposes only, it is not production ready. Use at your own risks.

Installation

Install Globally (Recommended for CLI)

After building the project (see below), you can install the CLI globally:

npm install -g simple-a2a-agent

This will make the simple-a2a-agent command available anywhere:

simple-a2a-agent --config samples/sample-config.json

Use with npx (No Install Required)

If you have published the package or want to use it directly from your local build, you can run:

npx simple-a2a-agent --config samples/sample-config.json

Or, if running from source before publishing, use:

npx tsx src/cli.ts --config samples/sample-config.json

Usage

Without config file

Export the required environment variables (depending on your model provider):

export OPENAI_API_KEY=sk-your-key

Run an agent (the description is the system prompt) :

npx simple-a2a-agent \
    --name "Cool agent" \
    --description "Reply with a cool description and emojis" \
    --llm-provider openai \
    --llm-model gpt-4o-mini

With config file

npx simple-a2a-agent --config samples/sample-config.json

or even shorter :

npx simple-a2a-agent -c samples/mcp-sample-config.json

Config file

{
  "agent": {
    "name": "Joke Agent",
    "description": "An agent that tells jokes and brings laughter to users.",
    "provider": { "organization": "ExampleOrg" },
    "version": "1.0.0",
    "skills": [
      {
        "id": "joke_telling",
        "name": "Joke Telling",
        "description": "Tells random jokes, puns, and humorous anecdotes on request.",
        "tags": ["joke", "humor", "fun"],
        "examples": [
          "Tell me a joke.",
          "Do you know any puns?",
          "Make me laugh!"
        ]
      }
    ]
  },
  "server": {
  },
  "llm": {
    "provider": {
      "type": "openai"
    },
    "model": "gpt-4o-mini"
  }
}

Adding MCP tools

{
    // ...
    "mcpServers": {
        "filesystem": {
        "type": "stdio",
        "command": "npx",
        "args": [
            "-y",
            "@modelcontextprotocol/server-filesystem",
            "/Users/neyric/Desktop"
        ],
        "env": { "NODE_ENV": "production" }
        }
    }
}

Cf https://voltagent.dev/docs/agents/mcp/

Improvements / TODO

• Remove option for config file (use first argument)
• Add an example for Anthropic Claude
• Split the agentCard description from the system prompt
• Use a future official A2A typescript SDK (for the time being, we imported files from https://github.com/google/A2A)
• Support for additional A2A features (streaming, input/output modes, ...)