Archived · v0.13.0 · Getting Started

Quickstart

Get a working agent and a client talking to each other in under five minutes.

1. Install

npm install @a2x/sdk @google/genai

Pick any LLM provider. This guide uses Google Gemini; swap in Anthropic or OpenAI any time (see Choose an LLM Provider).

2. Run your first agent

import { LlmAgent, toA2x } from '@a2x/sdk';
import { GoogleProvider } from '@a2x/sdk/google';
 
toA2x(
  new LlmAgent({
    name: 'hello_agent',
    description: 'A friendly assistant.',
    instruction: 'You are a helpful assistant. Keep answers short.',
    provider: new GoogleProvider({
      model: 'gemini-2.5-flash',
      apiKey: process.env.GOOGLE_API_KEY!,
    }),
  }),
  { port: 4000, defaultUrl: 'http://localhost:4000/a2a' },
);

Run it:

GOOGLE_API_KEY=... npx tsx server.ts

That's the whole server. Your agent now responds at:

GET http://localhost:4000/.well-known/agent.json — AgentCard for discovery.
POST http://localhost:4000/a2a — JSON-RPC endpoint.

3. Call it from another process

import { A2XClient } from '@a2x/sdk/client';
 
const client = new A2XClient('http://localhost:4000/.well-known/agent.json');
 
const task = await client.sendMessage({
  message: { role: 'user', parts: [{ text: 'Say hi in three words.' }] },
});
 
console.log(task.status.message?.parts);

That's it. You've shipped an A2A-compliant agent and called it from TypeScript.

Where to go next

If you want to…	Read
Customize the agent's behavior	Build an Agent
Give the agent tools it can call	Add Tools
Compose multiple agents	Multi-Agent Patterns
Mount the agent into Express/Next.js	Framework Integration
Stream responses token-by-token	Streaming Responses
Lock the agent behind auth	Authentication