Deploying an Agent with AWS AgentCore using CDK

This repo shows how to deploy a Fence or Strands agent on AWS AgentCore using AWS CDK instead of the CLI toolkit.

⚠️ CDK support for AgentCore is experimental and may change.

---

Why CDK?

CDK makes AgentCore deployments versioned, reproducible, and maintainable.
It handles Docker builds, ECR uploads, IAM, and CloudWatch automatically — perfect for scaling from local tests to production.

---

Agent Examples

Fence

from bedrock_agentcore.runtime import BedrockAgentCoreApp
from fence.agents.bedrock import BedrockAgent
from fence.models.bedrock import NovaPro

app = BedrockAgentCoreApp()
agent = BedrockAgent(
  identifier="editor_agent",
  model=NovaPro(region="eu-central-1", cross_region="eu"),
  description="Coach users to improve their writing.",
  tools=[], mcp_clients=[]
)

@app.entrypoint
def invoke(payload):
    user_message = payload.get("prompt", "Hello")
    return {"result": agent.run(user_message).answer}

Strands

from bedrock_agentcore.runtime import BedrockAgentCoreApp
from strands import Agent

app = BedrockAgentCoreApp()
agent = Agent(system_prompt="Coach users to improve writing.")

@app.entrypoint
def invoke(payload):
    user_message = payload.get("prompt", "Hello")
    return {"result": agent(user_message).message}

Structure

agentcore-cdk/
├── bin/
├── lib/
└── agents/
    ├── simple-agent-python-fence/
    └── simple-agent-python-strands/

Each agent has its own Dockerfile and requirements.txt. CDK builds and deploys them as AgentCore runtimes with IAM roles, ECR, and CloudWatch.

Deploy

npm install
npm run cdk bootstrap
npm run cdk deploy

After deployment, CDK outputs your runtime ARNs to invoke via Bedrock.