deploy-agentcore

<essential_principles> AWS Bedrock AgentCore is a serverless platform for AI agents at scale.

Architecture

AgentCore has 6 modular components:

• Runtime - Serverless hosting (direct_code_deploy or container)
• Gateway - Tool access via MCP (Lambda, OpenAPI, Smithy targets)
• Memory - STM (session) and LTM (persistent) storage
• Identity - Auth via IAM, Cognito, AWS JWT, external OAuth
• Observability - CloudWatch + OpenTelemetry tracing
• Policy - Cedar-based governance and authorization

Entry Point Pattern

All agents use BedrockAgentCoreApp with @app.entrypoint decorator:

from bedrock_agentcore import BedrockAgentCoreApp

app = BedrockAgentCoreApp()

@app.entrypoint
def invoke(payload: dict) -> dict:
    prompt = payload.get("prompt", "")
    result = your_agent_logic(prompt)
    return {"result": result}

if __name__ == "__main__":
    app.run()

Key CLI Commands

All commands: uv run agentcore [command]

Runtime: configure, deploy, invoke, status, destroy, stop-session Gateway: gateway create-mcp-gateway, gateway create-mcp-gateway-target Memory: memory create, memory list, memory status Identity: identity setup-cognito, identity setup-aws-jwt Policy: policy create-policy-engine, policy create-policy

See references/cli-reference.md for full command list.

Rules

• Agent names: underscores only (my_agent not my-agent)
• Never hardcode API keys - use Secrets Manager
• Windows: prefix with PYTHONIOENCODING=utf-8
• Memory mode order: STM_AND_LTM (not LTM_AND_STM) </essential_principles>

1. Deploy a new agent
2. Update existing deployment
3. Add Google OAuth
4. Create chat UI
5. Set up Gateway (MCP tools)
6. Configure Memory
7. Set up Identity/Auth
8. View logs/observability
9. Troubleshoot errors
10. Something else

Wait for response before proceeding.

After reading the workflow, follow it exactly.

<reference_index> All domain knowledge in references/:

• architecture.md - All AgentCore components explained
• cli-reference.md - Complete CLI command reference
• prerequisites.md - AWS setup, Python, uv requirements
• memory-modes.md - Memory configuration details
• common-errors.md - Error messages and fixes
• iam-policies.md - IAM role configuration </reference_index>

<workflows_index>

<templates_index>

<success_criteria> Deployment successful when:

• uv run agentcore deploy completes without errors
• uv run agentcore invoke returns expected response
• Agent handles sessions correctly
• External API keys work via Secrets Manager </success_criteria>