Strands Agents SDK

AWS Strands Agents SDK — callable agents, MCP, Bedrock-first, AgentCore.

AWS's open-source agent framework, already running inside Amazon Q Developer and AWS Glue internally. The leanest of the four by a real margin.

Core concepts

  1. The agent is callable: agent("...") both sends the message and runs the tool-calling loop — no separate Runner or .run() call needed. Strands Docs
  2. @tool decorator: Same lightweight registration pattern as the other frameworks, minimal ceremony.
  3. Bedrock-first, provider-neutral otherwise: Defaults to Amazon Bedrock with zero configuration; OpenAI, Anthropic direct, Ollama, LiteLLM, and Gemini are all supported.
  4. Native MCP support: Thousands of pre-built tools available through MCP servers with no custom integration code.
  5. AgentCore: A companion runtime for deploying Strands agents — positions the framework as the agent-building half of a two-piece AWS-native stack.

Why it matters

If you're on AWS with Bedrock and IAM, Strands cuts the glue code. Off AWS, another framework is usually simpler.

Where people get stuck

  • Broad IAM roles on tool permissions. An agent uses whatever you give it, so grant the minimum.
  • Assuming Strands replaces tracing and evals. It doesn't.
  • Choosing it with no AWS footprint.

Try this

Sketch an agent that calls one Bedrock model and one MCP tool, then list the IAM permission each call needs.

Interview question

When is Strands a good fit?

When you deploy on AWS, use Bedrock models, want MCP-compatible tools and prefer an AWS-supported SDK over your own boto3 loop.

Resources

YouTube learning

Core concepts from video walkthroughs:

Threads in the same domain as this chapter — go deeper on iHateReading without leaving the roadmap.

Python AI agent SDKs in 2026: OpenAI vs Pydantic-AI vs smolagents vs Strands cover

Python AI agent SDKs in 2026: OpenAI vs Pydantic-AI vs smolagents vs Strands

Same three-tool agent in four frameworks — pick by execution model and testing needs.

Agent harnessing: the part of building AI agents nobody talks about cover

Agent harnessing: the part of building AI agents nobody talks about

Loop, tools, memory, sandbox, and tracing live in your code — not the model weights.