AgentCore

This tutorial demonstrates how to deploy a Strands Agent to Amazon Bedrock AgentCore Runtime for production workloads.

Agent Architecture

Feature	Description
Deployment target	Amazon Bedrock AgentCore Runtime
Agent type	Single agent with Express server
Model	Claude Haiku 4.5 via Amazon Bedrock

Prerequisites

Node.js 20.x or later
AWS CLI configured with appropriate permissions
AWS account with AgentCore access
Docker with buildx support, or AWS CodeBuild permissions

Running Locally

cd typescript/02-deploy/01-agentcore
npm install
npm run dev

Test the agent:

curl http://localhost:8080/ping
curl -X POST http://localhost:8080/invocations \
  -H "Content-Type: application/json" \
  -d '{"input": {"prompt": "What is the weather?"}}'

Deploying to AgentCore

Setup Prerequisites

Run the setup script to create the IAM role and ECR repository:

chmod +x setup-prerequisites.sh
./setup-prerequisites.sh

Export the variables printed by the script:

export ROLE_ARN=<role-arn-from-script>
export REPO_URI=<repo-uri-from-script>
export AWS_REGION=us-east-1

Build and Push Docker Image

chmod +x build-image.sh
./build-image.sh

Note: The script auto-detects your environment and uses local Docker or AWS CodeBuild accordingly.

Create AgentCore Runtime

npm run deploy

Invoke Deployed Agent

AGENT_RUNTIME_ARN=<arn-from-deploy> npm run invoke "What is the weather?"

Key Concepts

AgentCore Runtime Service Contract

AgentCore Runtime requires a containerized application exposing two HTTP endpoints: /ping for status tracking and /invocations for processing requests:

app.get('/ping', (_req, res) => {
  res.json({ status: 'Healthy' });
});

app.post('/invocations', async (req, res) => {
  const prompt = req.body.input?.prompt;
  const result = await agent.invoke(prompt);
  res.type('text/plain').send(result.toString());
});

Creating an AgentCore Runtime

Use the AWS SDK to create an AgentCore Runtime that references your container image:

const command = new CreateAgentRuntimeCommand({
  agentRuntimeName: AGENT_NAME,
  agentRuntimeArtifact: {
    containerConfiguration: { containerUri: `${repositoryUri}:latest` }
  },
  networkConfiguration: { networkMode: 'PUBLIC' },
  roleArn
});

Invoking the AgentCore Runtime

Send requests to the deployed agent using the InvokeAgentRuntime API:

const command = new InvokeAgentRuntimeCommand({
  agentRuntimeArn: AGENT_RUNTIME_ARN,
  runtimeSessionId: sessionId,
  payload: new TextEncoder().encode(prompt),
  qualifier: 'DEFAULT'
});

Project Structure

├── Dockerfile               # Container image for AgentCore
├── setup-prerequisites.sh   # Setup script (IAM role, ECR repo)
├── build-image.sh           # Build script (Docker or CodeBuild)
├── package.json
├── tsconfig.json
└── src/
    ├── agent.ts             # Express server with Strands Agent
    ├── deploy-agent.ts      # AgentCore deployment script
    └── invoke-agent.ts      # Remote invocation script

Cleanup

# Delete AgentCore Runtime
aws bedrock-agentcore-control delete-agent-runtime \
  --agent-runtime-id <runtime-id> --region us-east-1

# Delete ECR repository
aws ecr delete-repository \
  --repository-name agentcore-deployment \
  --region us-east-1 --force

# Delete IAM role
aws iam delete-role-policy --role-name AgentCoreRuntimeRole --policy-name AgentCoreRuntimePolicy
aws iam delete-role --role-name AgentCoreRuntimeRole

# (Optional) Delete CodeBuild resources if created
aws codebuild delete-project --name agentcore-image-builder
aws iam delete-role-policy --role-name CodeBuildAgentCoreRole --policy-name CodeBuildPolicy
aws iam delete-role --role-name CodeBuildAgentCoreRole