Multi‑Agent Systems

/perspective/frameworks/claude-agent-sdk/

Frameworks · ReferenceHarness

Claude Agent SDK

Anthropic's agentic harness — the gather-context / act / verify loop behind Claude Code, with tools, subagents and MCP.

Scenario Autonomous coding & computer-use agents on ClaudeLicense Open source SDK (MIT)
01

Suitability

Best for

Building autonomous coding and computer-use agents on Claude with a production-tested agent loop.

Watch out

Anthropic / Claude-centric; tuned for the Claude model family.

02

Pattern support

ReActSupervisor
NativeComposable / supportingColor = ladder level L1–L4
Read on

Related frameworks

Closest options — by shared native patterns and same stack layer.

Harness

Deep Agents SDK

Opinionated, batteries-included harness on LangGraph: planning, subagents and a virtual file system out of the box.

Shares Supervisor · ReActView →Framework

AWS Strands SDK

Model-driven minimal SDK; the LLM weaves the graph from a 'bundle of threads' of tools.

Shares ReActView →Framework

CrewAI

Role-based 'Crews' of agents; flat abstraction over team-shaped workflows.

Shares SupervisorView →

Keep exploring

Compare the field.

ReferenceFrameworksSeventeen platforms across the stack, compared side by side.Compare the field ReferenceProtocolsHow agents interoperate — MCP, A2A and AG-UI on the wire.Compare protocols PatternsThe Six Coordination PatternsThe six patterns every architecture is built from.Browse the catalog

Search

Search patterns, frameworks, and pages.