Claude Agent SDK

The Claude Agent SDK is a toolkit for building custom AI agents on the same agent harness that powers Claude Code. It exposes the core agent loop — reading files, running shell commands, calling tools, and iterating against feedback — as a programmable library, so you can embed that behavior in your own applications instead of driving it through the terminal.

It is aimed at developers building agentic products: coding assistants, automation pipelines, internal developer tools, or any workflow where an agent needs to operate over a filesystem and a set of tools with minimal scaffolding.

Highlights

• Official Python and TypeScript SDKs that wrap the Claude Code agent loop
• Built-in file operations, command execution, and tool calling
• Extensible with custom tools and Model Context Protocol (MCP) servers
• Configurable permissions and system prompts to scope what an agent can do
• Supports streaming responses and multi-turn sessions

How it fits an AI-assisted workflow

Install the SDK and run a single-shot query:

import { query } from "@anthropic-ai/claude-agent-sdk";
 
for await (const message of query({
  prompt: "Summarize the open TODOs in this repository",
})) {
  console.log(message);
}

Use it when you want the agent's autonomy under your own control flow — wiring it into CI, a web backend, or a custom CLI — rather than interacting through Claude Code directly. You define the tools and permissions, and the SDK handles the loop.

Good to know