Google ADK for TypeScript

What does this agent do, and when should you use it?

Agent Development Kit (ADK) is Google’s open-source, code-first toolkit for TypeScript agent development. Developers define agent behavior, tool use, and orchestration in code; the supplied example uses LlmAgent, GOOGLE_SEARCH, and gemini-flash-latest. ADK supports sequential, parallel, loop, and routed multi-agent workflows, plus delegation to remote agents through the A2A protocol. It targets Node.js and browser ecosystems with ESM, CommonJS, and web bundles, TypeScript type safety, and Zod schema validation. The @google/adk-devtools package supplies commands including adk create, adk run, adk web, and adk deploy cloud_run; the supplied material does not define a fixed end-user output format or managed hosted service.

A developer creates an LlmAgent in a TypeScript file, setting fields such as name, description, model, instruction, and tools; the example adds GOOGLE_SEARCH to tools. From the agent project directory, npx adk run agent.ts starts an interactive CLI run, while npx adk web launches a development UI for testing and debugging. Agents can use the listed built-in Google Search, Google Maps, Vertex AI Search, and URL context tools, connect to MCP servers, wrap functions as tools, or add code execution. Developers can compose agents into sequential, parallel, loop, or routed workflows and delegate work to remote agents through A2A. The material does not specify a standardized response artifact, report, or message format.

1. A TypeScript developer prototyping a question-answering assistant that can use GOOGLE_SEARCH when needed can define an LlmAgent and test it with npx adk run agent.ts.
2. A team building an agent application for either Node.js or the browser can select from the documented ESM, CommonJS, and web bundles.
3. An engineering team that needs Google Search, Google Maps, Vertex AI Search, or URL context in an agent workflow can evaluate the listed built-in tools.
4. A developer coordinating specialized agents in sequential, parallel, loop, or routed flows can use ADK’s documented orchestration models.
5. An integration team that must delegate part of a workflow to a remote agent can evaluate the A2A protocol path.
6. A team that wants a browser-based debugging interface before deploying to Cloud Run can use adk web and assess adk deploy cloud_run.

What are this agent's strengths and limitations?

How do you install or deploy this agent?

Prerequisite: a current Node.js LTS release. In the agent project directory, install:

npm install @google/adk
npm install -D @google/adk-devtools

The supplied material does not specify credential setup for models, Google Search, Google Maps, Vertex AI Search, or other external services.

How do you use this agent?

Create agent.ts and import LlmAgent and GOOGLE_SEARCH:

import {LlmAgent, GOOGLE_SEARCH} from '@google/adk';

export const rootAgent = new LlmAgent({

name: 'search_assistant',
description: 'An assistant that can search the web.',
model: 'gemini-flash-latest',
instruction: 'You are a helpful assistant. Answer user questions using Google Search when needed.',
tools: [GOOGLE_SEARCH],
});

Run npx adk run agent.ts from that project directory for an interactive session, or run npx adk web for the development and debugging UI. The material also names adk create and adk deploy cloud_run, but does not provide their arguments, configuration files, or a complete deployment procedure.

FAQ