Anthropic takes the Claude Agent SDK out of beta
Anthropic has moved the Claude Agent SDK from beta to general availability. The SDK exposes the same agent loop, tool-use harness, and context-management machinery that powers Claude Code itself, packaged for developers building their own agents rather than using the CLI directly.
What's in the GA release
- A stable tool-definition schema, with backward-compatibility guarantees going forward
- Built-in support for subagents, background task execution, and structured output validation
- Production SLAs and rate-limit tiers separate from the standard Messages API
Before GA, teams building on the SDK were building against a moving target — tool schemas and agent-loop behavior could change between minor versions. A stable API surface is what actually unblocks shipping an agent-based feature into production, not just prototyping one.
Where this fits
This is the same SDK underlying Claude Code's own subagent and workflow features — a general-purpose agent, in other words, isn't a separate product built from scratch, it's the same harness with a different system prompt and tool set. For teams already comfortable calling the Messages API directly, the SDK's value is specifically the agent loop and context-compaction logic it handles for you, not new model capability.
Source: www.anthropic.com