Strands Agents

Type: full-code · Vendor: Strands Agents (AWS-originated, community-run) · Language: Python, TypeScript · License: Apache-2.0 · Status: active · Status in practice: emerging

Links: homepage docs repo

A model-driven approach to building AI agents in just a few lines of code.

Description. Lightweight, model-driven agent SDK: define an agent with a model, system prompt and tools, and let the model drive the loop. Native MCP server support, 13+ model providers (Amazon Bedrock default, plus OpenAI/Anthropic/Gemini/Ollama and others), multi-agent and autonomous-agent primitives, bidirectional audio streaming, Python-decorator tools and hot-reloadable tool directories. AWS-originated, now an open community project.

Agent loop shape. Model-driven loop: the model decides which tools to call and when to stop, given a model, system prompt and a set of decorator-defined tools. Supports multi-agent composition and bidirectional streaming for audio.

Primary use cases

building agents in a few lines with a model-driven loop
MCP-native tool integration across many providers
multi-agent and autonomous-agent systems, including real-time audio

flowchart TD GOAL[Task] --> LOOP[Model-driven loop] LOOP --> DECIDE{Call tool?} DECIDE -->|yes| TOOL[Decorator tool / MCP tool] TOOL --> LOOP DECIDE -->|done| OUT[Result] LOOP -.multi-agent.-> WORKERS[Sub-agents] LOOP -.audio.-> STREAM[Bidirectional streaming]

Key concepts

Model-driven loop (docs) — The model decides which tools to call and when to stop.
Tool decorators → tool-use — Python decorators register tools; tool directories hot-reload.
MCP → mcp — Native Model Context Protocol server support.