Think of a helpful assistant. You don't just ask it questions; you give it tasks. "Find the best flight to London for next Tuesday and book it using my saved credit card." To do this, the assistant needs to understand your request, use tools (a flight search engine, a payment system), and string together multiple steps to get the job done.

This is the idea behind an AI agent. It's more than a chatbot. An agent is an AI system that can understand a goal, make a plan, and use available tools to accomplish complex tasks. The Claude Agent SDK is a toolkit from Anthropic designed specifically to help developers build these powerful agents.

The SDK acts as the framework for your agent. It handles the complicated parts of interacting with Claude's large language models (LLMs), managing conversation history, and calling external tools. This lets you concentrate on defining what your agent should be able to do, rather than getting bogged down in the technical plumbing.

The SDK wasn't always called the Agent SDK. It started its life as the Claude Code SDK. Its initial focus was narrower: helping developers build applications that could generate and work with code. It was great for creating coding assistants or tools that could automate programming tasks.

However, the developers at Anthropic realized the underlying technology was capable of much more. The same logic that allows an AI to call a code interpreter can also be used to call a weather API, access a customer database, or interact with any other digital tool. The evolution to the Agent SDK reflects this bigger vision, shifting from a specialized coding tool to a general-purpose framework for building agents that can tackle a wide array of tasks.

The SDK provides a few key features that make agent development much simpler.

First is tool use, also known as function calling. You can define a set of tools in your code, like a function to get the current weather. The SDK makes these tools available to Claude. When a user asks, "What's the weather in San Francisco?", the model knows the get_weather tool exists and asks to run it with the correct input ("San Francisco"). The SDK manages this entire back-and-forth process.

Second, it handles the orchestration of complex, multi-step tasks. An agent can reason about a goal, break it down into smaller steps, use different tools for each step, and even handle errors along the way.

Finally, the SDK provides built-in conversation management. Agents need to remember what's been said previously to handle follow-up questions and maintain context. The SDK takes care of this automatically.

Getting the Claude Agent SDK running is straightforward. It currently supports two popular programming languages: TypeScript and Python. You'll need the standard package manager for your chosen language.

For TypeScript, you'll use npm (Node Package Manager). Make sure you have Node.js installed, then run this command in your project's terminal:

For Python, you'll use pip, Python's package installer. With Python already set up, you can install the SDK with this command:

Once the package is installed, you'll also need an API key from Anthropic to authenticate your requests. With these pieces in place, you're ready to start building your first agent.