Claude Agent SDK: Build AI Agents with Anthropic’s Framework

Bottom line up front: The Claude Agent SDK is Anthropic’s open-source framework for building AI agents — systems that can use tools, browse the web, run code, and complete multi-step tasks autonomously. It’s the building block for anyone who wants to create custom AI-powered applications using Claude.

Key Takeaways

• The Claude Agent SDK is Anthropic’s open-source Python framework for building AI agents.
• Agents built with the SDK can use tools, search the web, run code, and complete multi-step tasks.
• Designed for developers — requires Python knowledge but has excellent documentation and examples.
• Works with Claude Sonnet 4.6 and Opus 4.6 via the Anthropic API.
• Used to build custom AI assistants, automated workflows, research agents, and production applications.

What Is the Claude Agent SDK?

The Claude Agent SDK (sometimes called the Claude Code SDK or Anthropic Agent Framework) is an open-source Python library published by Anthropic. It gives developers the tools to build AI agents — programs where Claude doesn’t just answer questions but actually takes actions, uses external tools, and works through tasks autonomously over multiple steps.

The difference between a standard Claude API call and an agent built with the SDK is significant. A basic API call sends a message and gets a response. An agent can receive a goal, decide what tools to use, call those tools, process the results, decide what to do next, and keep going until the task is complete — all without human input at every step.

This opens up a completely different class of applications. Instead of “Claude answers my question,” you get “Claude completes my research project.” Instead of “Claude writes a draft,” you get “Claude researches the topic, outlines the piece, writes the draft, and checks it against my style guide.” The SDK is what makes that level of automation possible.

Key Concepts in the Agent SDK

Tools

Tools are the actions an agent can take beyond just generating text. The SDK lets you define custom tools — functions that Claude can call to do things in the real world. Built-in tools include web search, code execution, and file system access. Custom tools can be anything your Python code can do: query a database, call an API, read a document, send an email, update a spreadsheet.

When Claude decides to use a tool, it generates a structured tool call that your code executes. The result comes back to Claude, which then decides what to do with it. This loop continues until Claude completes the task or hits a stopping condition you define.

The Agent Loop

The agent loop is the core of how agents work. Here’s the basic cycle:

1. You give the agent a task and a set of available tools.
2. Claude decides whether to use a tool or respond directly.
3. If using a tool, Claude generates a tool call with the right parameters.
4. Your code runs the tool and returns the result to Claude.
5. Claude processes the result and decides whether to use another tool or finish.
6. The loop continues until the task is done or a limit is reached.

The SDK handles most of this loop management for you. You define the tools, set the model, and provide the task. The SDK handles the back-and-forth between Claude and your tools automatically.

Multi-Agent Systems

The SDK also supports multi-agent architectures — systems where multiple Claude instances work together. An orchestrator agent breaks down a complex task and assigns subtasks to specialized subagents. Each subagent works on its part and returns results to the orchestrator, which assembles the final output. This is how you build agents that can tackle very large, complex tasks that would exceed a single agent’s context window or capabilities.

What Can You Build With the Agent SDK?

The range of applications is broad. Here are real examples of what people build:

• Research agents: Search the web, read papers, synthesize findings, and produce a formatted research report — fully automated.
• Data pipeline agents: Pull data from APIs, clean it, analyze it, and write the results to a database or spreadsheet.
• Code review agents: Read a GitHub PR, analyze the changes, run test suites, and write a detailed review comment.
• Customer support agents: Answer customer questions by looking up account data, checking policy documents, and escalating when needed.
• Content production agents: Research a topic, draft an article, check it against a style guide, and publish it to a CMS — all in one automated run.
• Sales and outreach agents: Research prospects, draft personalized outreach emails, and log activities in a CRM.

Getting Started With the Agent SDK

The SDK is a Python package. Here’s the basic setup:

pip install anthropic

# Basic agent with a custom tool
import anthropic

client = anthropic.Anthropic()

tools = [
    {
        "name": "get_weather",
        "description": "Get the current weather for a city",
        "input_schema": {
            "type": "object",
            "properties": {
                "city": {"type": "string", "description": "City name"}
            },
            "required": ["city"]
        }
    }
]

response = client.messages.create(
    model="claude-sonnet-4-6-20250514",
    max_tokens=1024,
    tools=tools,
    messages=[{"role": "user", "content": "What's the weather in London?"}]
)

This is a simplified example. The full SDK includes utilities for managing conversation history, handling tool results, managing agent loops, and building multi-agent pipelines. The documentation on Anthropic’s website walks through each component in detail.

Who Should Use the Agent SDK?

The Agent SDK is built for developers and technical teams. You need to be comfortable with Python to use it effectively. You also need an Anthropic API key and understand the basics of how Claude works. If you’re not a developer, you probably want to use Claude.ai or one of the built-in integrations instead.

For developers, the SDK is the right tool if you want to:

• Build a custom AI assistant for your specific workflow or business.
• Automate multi-step tasks that currently require human decision-making.
• Integrate Claude into an existing application or internal tool.
• Build and ship a product that uses Claude as the AI backbone.

Developers who want a complete AI coding environment should also explore Claude Code, which is itself built on top of the Agent SDK and gives you an AI pair programmer in your terminal. For how developers broadly are integrating AI into their work, see our AI for developers guide.

Agent SDK vs. Claude.ai vs. Claude API

These are three different ways to use Claude, for three different purposes:

• Claude.ai: The consumer interface. Great for individuals using Claude for everyday tasks. No coding required. See our Claude beginner’s guide.
• Anthropic API: Direct programmatic access to Claude. You send a message, you get a response. Simple and flexible, but you build the surrounding logic yourself.
• Agent SDK: A framework on top of the API. It provides the structure, tooling, and patterns for building agents that take autonomous action. The right choice when you need more than a simple request-response loop.

Business Applications by Industry

The Agent SDK is being used across many industries to automate knowledge work:

• Marketing teams: Agents that research competitors, draft content, and analyze campaign performance. See how AI is reshaping marketing.
• Accounting firms: Agents that extract data from financial documents, reconcile accounts, and flag anomalies. Read about AI in accounting.
• Legal teams: Agents that review contracts, identify risk clauses, and research case law. Learn how law firms use AI.
• HR departments: Agents that screen resumes, draft job postings, and answer employee questions. Explore AI tools for HR.

Choosing the Right Claude Model for Your Agent

When building agents with the SDK, model selection matters. Use Claude Sonnet 4.6 for most tasks — it’s fast, capable, and cost-effective for high-volume agent workflows. Use Claude Opus 4.6 for the orchestrator in complex multi-agent systems or for steps that require the deepest reasoning. Use extended thinking for specific steps in your agent pipeline where analytical depth is critical. For plan options, see our Claude plans guide.

External Resources

For technical background on AI agent architectures, read the Wikipedia overview of AI agents. Anthropic’s official tool use documentation is the authoritative technical reference. For foundational research on tool-using language models, see the arXiv paper on Toolformer, which laid key groundwork for this area.

Free Resource: Agent SDK Starter Kit

We’ve created a free Agent SDK Starter Kit with five ready-to-run agent templates — research agent, data agent, content agent, support agent, and email agent. Download it free in our products library and start building your first AI agent today. Get the Agent SDK Starter Kit →

Stay Updated

The Agent SDK is actively developed. New capabilities, better tool support, and improved multi-agent patterns are released regularly. Subscribe to the Beginners in AI newsletter to stay current on what’s new in the Claude developer ecosystem. Join the newsletter →

Frequently Asked Questions

Do I need to know Python to use the Claude Agent SDK?

Yes. The SDK is a Python package and requires comfortable Python knowledge to use effectively. If you’re not a developer, Claude.ai’s built-in tools (Chrome extension, desktop app, Office integrations) are better fits for your needs.

Is the Claude Agent SDK free?

The SDK itself is free and open source. You pay for the Anthropic API usage — Claude charges per token, so costs depend on how much your agents use Claude. For personal projects, costs are usually very low. For production systems, you’ll want to model expected usage to estimate costs.

What’s the difference between the Agent SDK and LangChain?

LangChain is a third-party framework that supports multiple AI providers including Claude. The Claude Agent SDK is Anthropic’s own first-party framework, built specifically for Claude. It has tighter integration with Claude’s capabilities (including extended thinking and tool use patterns) but doesn’t support other AI providers. Choose the Agent SDK if you’re committed to building on Claude; consider LangChain if you need provider flexibility.

Can I build a commercial product with the Agent SDK?

Yes. The SDK is open source and Anthropic’s API terms allow commercial use subject to their usage policies. Review Anthropic’s Terms of Service for the specifics on commercial applications.

You May Also Like

• What Are Agent Handoffs?
• The Future of AI Agents: What’s Coming in 2026-2027
• AI Agent Security: Risks, Guardrails, and Best Practices
• Best AI Agent Tools for Non-Developers
• How to Use AI: A Step-by-Step Guide

---

Get free AI tips daily → Subscribe to Beginners in AI

Sources

This article draws on official documentation, product pages, and industry reporting. Specific sources are linked inline throughout the text.

Last reviewed: April 2026