AWS Agentcore Agent Workflow

Creating AI Agents on AWS Bedrock with AgentCore

Complete workflow for creating and deploying AI agents using AWS Bedrock AgentCore services via MCP servers.

Prerequisites

Before starting, verify:

• MCP servers configured (required):
  • Bedrock AgentCore MCP Server for documentation and guides
  • Full AWS MCP Server for API operations
• AWS credentials configured (via MCP)
• AWS account has Bedrock AgentCore access enabled
• Required IAM permissions for AgentCore operations
• Target AWS region selected (e.g., us-west-2)
• Bedrock model access configured (if using specific models)

Verify MCP servers are available:

Look for these MCP tools:

• mcp__*bedrock-agentcore*__* or mcp__*awslabs.amazon-bedrock-agentcore-mcp-server*__* → Bedrock AgentCore MCP Server
• mcp__aws-mcp__aws___call_aws → Full AWS MCP Server (for API calls)

Note: This workflow requires MCP servers. If MCP servers are not configured, use the aws-mcp-setup skill to set them up.

Core Workflow

Step 1: Create Runtime

Runtime provides the serverless execution environment for your agent.

1. Get Runtime deployment guide using Bedrock AgentCore MCP Server:

   • Use mcp__*bedrock-agentcore*__manage_agentcore_runtime or mcp__*awslabs.amazon-bedrock-agentcore-mcp-server*__manage_agentcore_runtime to get comprehensive Runtime deployment instructions

2. Create runtime using AWS MCP Server:

   • Use mcp__aws-mcp__aws___call_aws with service bedrock-agentcore-control, operation CreateRuntime
   • Required parameters: name (runtime name), region (AWS region)

3. Save the runtime identifier from the response for use in subsequent steps.

4. Verify runtime:

   • Use mcp__aws-mcp__aws___call_aws with service bedrock-agentcore-control, operation GetRuntime, and the runtime identifier

Step 2: Create Agent

Create the AI agent with your chosen foundation model.

1. Create agent using AWS MCP Server:

   • Use mcp__aws-mcp__aws___call_aws with service bedrock-agentcore-control, operation CreateAgent
   • Required parameters: runtimeIdentifier, name, agentConfiguration (with modelId), region
   • Optional: temperature, maxTokens in agentConfiguration

2. Save the agent identifier from the response for use in subsequent steps.

3. Verify agent:

   • Use mcp__aws-mcp__aws___call_aws with service bedrock-agentcore-control, operation GetAgent, runtime identifier, and agent identifier

Step 3: Test Basic Agent

Invoke the agent to verify it's working:

• Use mcp__aws-mcp__aws___call_aws with service bedrock-agentcore-runtime, operation InvokeAgent
• Required parameters: runtimeIdentifier, agentIdentifier, input (with messages array), region
• Example message: {"role": "user", "content": "Hello, can you introduce yourself?"}

Optional Services

After creating your basic agent, you can enhance it with optional services:

• Memory: Persistent conversation memory across sessions
• Gateway: Connect external REST APIs as tools
• Code Interpreter: Secure Python code execution
• Browser: Web automation and scraping
• Observability: Monitoring, logging, and tracing

For detailed setup instructions: See references/optional-services.md

Workflow Checklist

Use this checklist to track your agent creation progress:

Agent Creation Progress:
- [ ] MCP servers configured (Bedrock AgentCore MCP Server, AWS MCP Server)
- [ ] Prerequisites verified (AWS credentials via MCP, permissions, region)
- [ ] Runtime created and verified
- [ ] Agent created with foundation model
- [ ] Basic agent invocation tested
- [ ] Memory service configured (if needed)
- [ ] Gateway targets deployed (if needed)
- [ ] Code Interpreter enabled (if needed)
- [ ] Browser service enabled (if needed)
- [ ] Observability configured
- [ ] Agent tested with all enabled services
- [ ] IAM permissions verified for production
- [ ] CloudWatch alarms configured (optional)

Common Configurations

Quick reference for common agent setups:

• Basic Conversational Agent: Runtime + Agent only
• Agent with Memory: Runtime + Agent + Memory
• Full-Featured Agent: Runtime + Agent + Memory + Code Interpreter + Browser + Observability

For detailed configuration examples: See references/common-configurations.md

Testing

After creating your agent, test it to ensure everything works correctly.

For comprehensive testing procedures: See references/testing.md

Troubleshooting

If you encounter issues during agent creation or operation:

For troubleshooting guide: See references/troubleshooting.md

Additional Resources

• MCP Server Setup: Use aws-mcp-setup skill to configure Bedrock AgentCore MCP Server and AWS MCP Server
• MCP Documentation Tools:
  • mcp__*bedrock-agentcore*__search_agentcore_docs - Search AgentCore documentation
  • mcp__*bedrock-agentcore*__fetch_agentcore_doc - Fetch complete documentation pages
  • mcp__*bedrock-agentcore*__manage_agentcore_runtime - Runtime deployment guide
  • mcp__*bedrock-agentcore*__manage_agentcore_memory - Memory management guide
  • mcp__*bedrock-agentcore*__manage_agentcore_gateway - Gateway deployment guide
• Service-specific guides: See ../aws-agentcore/references/ for detailed documentation
• Credential management: ../aws-agentcore/references/credential-management.md
• AWS Documentation: Amazon Bedrock AgentCore