A2A Server

The A2A Server exposes your Agentium agents as standard A2A endpoints. Any A2A-compatible client — including agents built with LangGraph, CrewAI, AutoGen, or other frameworks — can discover and interact with your agents.

Setup

import express from "express";
import { Agent, openai, defineTool } from "@agentium/core";
import { createA2AServer } from "@agentium/transport";
import { z } from "zod";

const app = express();

const assistant = new Agent({
  name: "assistant",
  model: openai(process.env.OPENAI_MODEL ?? "gpt-6.1-sol"),
  instructions: "You are a helpful assistant.",
});

const calculator = new Agent({
  name: "calculator",
  model: openai(process.env.OPENAI_FAST_MODEL ?? "gpt-6-luna"),
  instructions: "You are a calculator. Use the calculate tool.",
  tools: [
    defineTool({
      name: "calculate",
      description: "Evaluate a math expression",
      parameters: z.object({
        expression: z.string(),
      }),
      execute: async (args) => String(eval(args.expression)),
    }),
  ],
});

createA2AServer(app, {
  agents: { assistant, calculator },
  basePath: "/",
  provider: {
    organization: "MyCompany",
    url: "https://mycompany.com",
  },
});

app.listen(3001, () => {
  console.log("A2A Server running at http://localhost:3001");
});

Agent Card

The server automatically generates an Agent Card at /.well-known/agent.json:

{
  "name": "Agentium Agent Server",
  "description": "Multi-agent server with 2 agents: assistant, calculator",
  "url": "/",
  "version": "1.0.0",
  "provider": {
    "organization": "MyCompany",
    "url": "https://mycompany.com"
  },
  "capabilities": {
    "streaming": true,
    "pushNotifications": false,
    "stateTransitionHistory": true
  },
  "skills": [
    { "id": "assistant", "name": "assistant", "description": "..." },
    { "id": "calculator", "name": "calculator", "description": "..." }
  ]
}

Each agent is listed as a skill. The card is auto-generated from agent metadata — no manual configuration needed.

JSON-RPC Methods

The server handles these A2A methods at the basePath:

message/send

Synchronous request/response. Runs the agent and returns the complete result.

curl -X POST http://localhost:3001/ \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "message/send",
    "params": {
      "message": {
        "role": "user",
        "parts": [{ "kind": "text", "text": "Hello!" }],
        "metadata": { "agentName": "assistant" }
      }
    }
  }'

message/stream

Server-Sent Events (SSE) streaming. Returns chunks as the agent generates output.

curl -X POST http://localhost:3001/ \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "message/stream",
    "params": {
      "message": {
        "role": "user",
        "parts": [{ "kind": "text", "text": "Tell me a story" }]
      }
    }
  }'

tasks/get

Retrieve a task by ID, optionally limiting history length.

curl -X POST http://localhost:3001/ \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 2,
    "method": "tasks/get",
    "params": { "id": "task-uuid-here" }
  }'

tasks/cancel

Cancel a running task.

curl -X POST http://localhost:3001/ \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 3,
    "method": "tasks/cancel",
    "params": { "id": "task-uuid-here" }
  }'

Agent Routing

When multiple agents are registered, the server resolves which agent to use:

Explicit — Set metadata.agentName in the message to target a specific agent
Name matching — The server checks if the message text mentions an agent name
Default — Falls back to the first registered agent

Examples

# Explicit routing via metadata
curl -X POST http://localhost:3001/ \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "message/send",
    "params": {
      "message": {
        "role": "user",
        "parts": [{ "kind": "text", "text": "What is 2 + 2?" }],
        "metadata": { "agentName": "calculator" }
      }
    }
  }'

Use A2ARemoteAgent from @agentium/core for the legacy client; see client configuration. Use A2AV1Client with the separate A2A 1.0 server.

Task Lifecycle

Each request creates a task that transitions through these states:

submitted → working → completed
                   → failed
                   → canceled

The task stores the full message history and any artifacts (tool results).

Configuration

Record<string, Agent>

required

Map of agent name to Agent instance.

string

default:"/"

Base path for the JSON-RPC endpoint.

object

Provider information included in the Agent Card.

organization (string) — Organization name
url (string) — Organization URL

string

default:"1.0.0"

Version string for the Agent Card.

Find an integration

Voice and call adapters

Sandbox and telemetry

Providers and services

Protocols and devices

Setup

Agent Card

JSON-RPC Methods

message/send

message/stream

tasks/get

tasks/cancel

Agent Routing

Examples

Task Lifecycle

Configuration

​Setup

​Agent Card

​JSON-RPC Methods

​message/send

​message/stream

​tasks/get

​tasks/cancel

​Agent Routing

​Examples

​Task Lifecycle

​Configuration

Setup

Agent Card

JSON-RPC Methods

message/send

message/stream

tasks/get

tasks/cancel

Agent Routing

Examples

Task Lifecycle

Configuration