Build, deploy, and operate AI agents using Strands SDK, MCP, and Amazon Bedrock AgentCore.
Open-source Python/TypeScript SDK for building AI agents. Model-driven approach: the LLM decides what to do, not fixed code logic.
One agent with focused role, clear tools, straightforward prompts. Start here.
Wrap agents as callable tools. Parent agent invokes specialized child agents.
Parallel execution (swarms) or DAG-based workflows (graphs) for complex tasks.
A2A protocol for seamless delegation between agents with context transfer.
An open standard for AI systems to communicate with tools, data sources, and services. Build tools once, integrate anywhere.
Callable actions with JSON schemas
Read-only persistent data
Reusable instruction templates
Fully managed platform for deploying and operating AI agents at enterprise scale. Framework-agnostic, model-agnostic.
Serverless execution
Session state
Tool mediation
Access control
OpenTelemetry
The sample app includes a complete, working agent you can run in under 5 minutes.
# 1. Start the task API
cd sample-app && npm install && npm run dev
# 2. Install agent dependencies
cd agent && pip install -e ".[dev]"
# 3. Run the demo (no AWS needed)
python scripts/run-demo.py --mock
# 4. Or run interactively
python scripts/run-agent.py --mock