Claude Agent SDK 101: how it differs from Claude Code, and when to use which

CLI vs embedding in your own app — including the recent rename.

Same engine, two interfaces

Claude Code wraps the engine in an interactive CLI/editor experience. Claude Agent SDK exposes the same engine as a TypeScript/Python library to embed in your own apps, servers or pipelines.

The March 2026 rename

Formerly "Claude Code SDK", officially renamed to "Claude Agent SDK" in March 2026. Old packages (`@anthropic-ai/claude-code`, `claude-code-sdk`) are deprecated; current docs/dev live under the new names.

Installing it

TypeScript via npm, Python via pip (3.10+):

# TypeScript
$ npm i @anthropic-ai/claude-agent-sdk

# Python (3.10+)
$ pip install claude-agent-sdk

Same agent loop, tools and context management

You call the same agent loop programmatically instead of driving it via the CLI — good for CI automated review, embedding in internal systems, or custom workflow automation.

Which to pick

Claude Code for day-to-day interactive work; Claude Agent SDK for embedding agent behavior into your own service or CI pipeline. Many teams use both.

Relation to ConfigDeck

ConfigDeck's configs (subagents/skills/commands) target Claude Code's `.claude/` format. Building with the Agent SDK means implementing agent behavior directly in your app's code instead.

FAQ

What was Claude Agent SDK called before?
"Claude Code SDK" — officially renamed in March 2026.
How do I install it?
`npm i @anthropic-ai/claude-agent-sdk` (TS) or `pip install claude-agent-sdk` (Python 3.10+).
Claude Code or Claude Agent SDK?
Claude Code for interactive daily work; Agent SDK for embedding into your app or CI.
Do ConfigDeck configs work with the Agent SDK?
No — ConfigDeck targets Claude Code's .claude/ format; the SDK means writing agent behavior directly in your code.