Anthropic Launches Claude Agent SDK — Now Everyone Can Build AI Agents
Anthropic just made a major announcement: Claude Code SDK has been renamed to Claude Agent SDK. And this isn't just rebranding — it's a full-fledged platform for creating AI agents of any type.
What is this all about?
Remember Claude Code — the developer tool that works right in your terminal? Turns out it can be used not just for coding, but for deep research, video creation, note-taking, and tons of other tasks. The Anthropic team realized they'd built a universal agent engine and decided to open it up to everyone.
What kinds of agents can you build?
The SDK enables you to build finance agents for investment analysis, personal assistants for booking travel and managing calendars, customer support agents, and deep research agents for large document collections.
Basically, you can automate any workflow.
How does it work?
Agents operate in a cycle: gather context → take action → verify work → repeat. The SDK provides all the necessary tools to implement this cycle.
For example, if you're building an email agent:
- File system stores information that can be loaded into the model's context
- Subagents help process tasks in parallel and manage context
- Tools provide access to file operations, code execution, and web search
Subagents are especially useful: they allow running multiple tasks simultaneously and use isolated context windows, sending back only relevant information.
What's included in the SDK?
Automatic context management so the agent doesn't hit limits, a rich tool ecosystem for file operations, code execution, web search, and MCP extensibility, advanced permissions system with fine-grained control over agent capabilities, built-in error handling, session management, and monitoring.
MCP Servers for Your Agents
By the way, if you want to extend your Claude agents' capabilities, there are plenty of ready-made MCP servers available at fastmcp.me. It's a great way to quickly add new features to your agent without having to build everything from scratch.
Usage Examples
Anthropic has already released a demo email agent on GitHub that can:
- Search emails using natural language
- Help compose replies with Claude's assistance
- Summarize email threads
- Extract information
How to get started?
The SDK is available for Python. The simplest query looks like this: async for message in query(prompt="What is 2 + 2?").
For more complex scenarios, there's ClaudeSDKClient with support for custom tools and hooks. You can define your own Python functions that Claude will invoke as needed.
Why does this matter?
This is a huge step toward a virtual collaborator. Claude 4 already shows impressive results on agent tasks — the model can maintain full context, focus on long-term projects, and write key information to local files to improve its performance.
Now these capabilities are available to any developer through an open SDK.
Bottom line: Claude Agent SDK isn't just a coding tool. It's a full-featured platform for building intelligent agents that can automate practically any task. And the best part — you don't need to reinvent the wheel, you can use ready-made MCP servers from fastmcp.me and immediately get a powerful assistant.
Time to experiment 🚀