Anthropic launches the Claude Agent SDK

Anthropic renamed the Claude Code SDK to the Claude Agent SDK and repositioned it as infrastructure for building general-purpose autonomous agents, not just coding assistants. The toolkit, which had powered Claude Code internally, gave developers the same primitives: file-system access so an agent can read and write its own working context, a bash tool for general computer use, support for custom tools, parallel “subagents” with isolated context windows, and built-in Model Context Protocol connections to external services such as Slack and GitHub.

The rename reflected a bet that the hard parts of building a coding agent — managing long contexts without losing track of state, giving a model reliable access to a file system and a shell, coordinating subtasks — were the same hard parts involved in building agents for other domains, such as finance research, customer support or personal assistance. Anthropic pointed to automatic context compaction, which summarises and discards old conversation history to keep an agent within its context window during long-running tasks, as a capability developers building agents from scratch would otherwise have to solve themselves.

The move followed a broader pattern of frontier labs converting internal tooling into public developer products once it had been proven on their own flagship application: Claude Code itself, and later Anthropic’s Agent Skills format for packaging reusable agent instructions, both went through the same path from internal use to general release.