Anthropic
Claude Agent SDK
Library that exposes the Claude Code agent loop, built-in tools and context management for building agents in Python and TypeScript.
Install
Run in your terminal
pip install claude-agent-sdk
Run in your terminal
npm install @anthropic-ai/claude-agent-sdk
Latest version read from pypi package claude-agent-sdk.
Multi-agent patterns
Handoffs
One agent passes the conversation to another agent, which takes over.
Not shown in the docs.
Subagents / agents as tools
A main agent calls other agents like tools and gets their result back, each with its own context.
How Claude Agent SDK does itGraph or workflow
You define steps and edges (or a workflow), and the framework runs agents and functions through them.
Not shown in the docs.
Group chat / crew
Several agents with roles work together on a task, coordinated by a manager or a selection rule.
Not shown in the docs.
Human in the loop
The run pauses for a person to approve, edit or answer before it continues.
How Claude Agent SDK does itProtocols and hosting
Hosted runtime
Self-hosted only (none documented)
Questions
Does Claude Agent SDK support MCP?
Yes. Claude Agent SDK's documentation shows agents using tools from MCP servers.
Does Claude Agent SDK support the A2A protocol?
Its documentation does not cover A2A as of 2026-10-07.
What multi-agent patterns does Claude Agent SDK support?
The docs show: subagents / agents as tools, human in the loop.
Links
Alternatives in the same languages
Get the weekly agent stack update
New official MCP servers, spec changes and harness releases, checked against the source. One email a week, no fluff.
Reviewed Oct 7, 2026.