Agent SDK

The Claude Agent SDK exposes Claude Code’s full agentic engine as a programmable library. Available in Python (claude-agent-sdk) and TypeScript (@anthropic-ai/claude-agent-sdk).

What It Is

The same infrastructure that powers Claude Code, but programmable. Instead of interacting through a terminal, you build applications that use Claude’s agent loop: reading files, running commands, editing code, searching the web - all orchestrated by the model.

Two Levels

There’s an important distinction between the Client SDK and the Agent SDK:

The Agent Loop

Same loop as Claude Code:

prompt → Claude thinks → tool call → execute → observe → repeat → final response

The SDK handles all of this. You provide a prompt, and query() streams back messages as Claude works through the task.

Core API

Python

from claude_agent_sdk import query

async for message in query(
    prompt="Find all TODO comments in this project and create a summary",
    options={
        "model": "claude-sonnet-4-6",
        "permissionMode": "default",
    }
):
    if message.type == "text":
        print(message.text, end="")
    elif message.type == "tool_use":
        print(f"Using tool: {message.tool}")

TypeScript

import { query } from "@anthropic-ai/claude-agent-sdk";

for await (const message of query({
  prompt: "Find all TODO comments in this project and create a summary",
  options: {
    model: "claude-sonnet-4-6",
    permissionMode: "default",
  },
})) {
  if (message.type === "text") {
    process.stdout.write(message.text);
  }
}

Built-in Tools

The Agent SDK comes with the same tools as Claude Code:

Custom Tools via MCP

Add custom tools by creating in-process MCP servers:

from claude_agent_sdk import query, create_sdk_mcp_server

server = create_sdk_mcp_server("my-tools")

@server.tool
def lookup_user(user_id: str) -> str:
    """Look up a user by their ID."""
    return db.get_user(user_id).to_json()

async for msg in query(
    prompt="Look up user 123 and summarize their recent activity",
    options={"mcpServers": [server]},
):
    print(msg)

Subagents

Define specialized agents with isolated context, restricted tools, and model overrides:

from claude_agent_sdk import query, AgentDefinition

security_agent = AgentDefinition(
    name="security-reviewer",
    instructions="Review code for security vulnerabilities. Focus on injection, auth bypass, and data leaks.",
    model="claude-opus-4-6",
    tools=["Read", "Grep", "Glob"],
)

async for msg in query(
    prompt="Review the auth module for security issues",
    options={"agents": [security_agent]},
):
    print(msg)

Non-Coding Uses

The Agent SDK isn’t limited to coding tasks:

• Deep research - search the web, read pages, synthesize findings
• Content creation - generate documents from templates and data
• Data processing - read files, transform data, write outputs
• Note-taking - capture and organize information from multiple sources

Key Features

Next

• Agent SDK quickstart - build a file-processing agent
• Claude Code features - the features that the SDK exposes programmatically