AgentCore CLI in Practice — Build a Tech Trend Advisor with Key Features Combined

Introduction

This series has covered the four main AgentCore CLI features individually:

• Part 1 — Runtime (basic lifecycle)
• Part 2 — Memory (conversation persistence)
• Part 3 — Gateway (external MCP server)
• Part 4 — Evaluations (quality measurement)

In this bonus finale, we combine these four key features into a single project: a "tech trend advisor" that remembers the user's expertise (Memory), searches the web for latest trends (Gateway), and auto-measures response quality (Evaluations).

AgentCore CLI is in Public Preview (v0.3.0-preview). Commands, options, and generated templates may change before GA. This article reflects behavior as of March 2026.

Prerequisites

• Environment from Part 1 (Node.js 20+, uv, AWS CLI, AgentCore CLI v0.3.0-preview)
• us-east-1 region — Evaluator CloudFormation resource is not supported in ap-northeast-1 (see Part 4)

If AWS_REGION is set in your environment, it may override the region in aws-targets.json. If it's set to something other than us-east-1, run export AWS_REGION=us-east-1 or unset AWS_REGION before proceeding.

What We're Building

Project Setup

As learned in Part 3, Gateway requires --no-agent first, then adding the Gateway before the agent.

Steps 1–3: Project Creation and Gateway Setup

agentcore create --name AgentCoreFull --no-agent --skip-git
cd AgentCoreFull
 
# Add Gateway
agentcore add gateway --name my-gateway
 
# Add Exa AI MCP server as target
agentcore add gateway-target \
  --type mcp-server \
  --name exa-search \
  --endpoint https://mcp.exa.ai/mcp \
  --gateway my-gateway

Step 4: Add Agent (Memory + Gateway Integration)

With --memory longAndShortTerm, both the Gateway client and Memory session manager are auto-generated.

agentcore add agent \
  --name TechAdvisor \
  --framework Strands \
  --model-provider Bedrock \
  --language Python \
  --memory longAndShortTerm

Step 5: Add Evaluator

agentcore add evaluator \
  --name ResponseQuality \
  --level SESSION \
  --model us.anthropic.claude-sonnet-4-5-20250929-v1:0 \
  --instructions "Evaluate the overall quality of the agent's response. Consider accuracy, helpfulness, personalization based on user context, and clarity. Context: {context}" \
  --rating-scale 1-5-quality

Customize the System Prompt

Edit the auto-generated main.py to set the tech trend advisor persona.

system_prompt="""
    You are a personalized tech trend advisor. You remember the user's
    technical background, skills, and interests from previous conversations.
    When asked about trends or recommendations, search the web for the
    latest information and tailor your advice based on what you know about
    the user. Always be specific and actionable.
""",

The full main.py combines the Memory integration from Part 2 and the Gateway client from Part 3. It uses get_all_gateway_mcp_clients() for Gateway tools and get_memory_session_manager() for session management.

from strands import Agent, tool
from bedrock_agentcore.runtime import BedrockAgentCoreApp
from model.load import load_model
from mcp_client.client import get_all_gateway_mcp_clients
from memory.session import get_memory_session_manager
 
app = BedrockAgentCoreApp()
log = app.logger
 
mcp_clients = get_all_gateway_mcp_clients()
 
tools = []
 
@tool
def add_numbers(a: int, b: int) -> int:
    """Return the sum of two numbers"""
    return a+b
tools.append(add_numbers)
 
for mcp_client in mcp_clients:
    if mcp_client:
        tools.append(mcp_client)
 
 
def agent_factory():
    cache = {}
    def get_or_create_agent(session_id, user_id):
        key = f"{session_id}/{user_id}"
        if key not in cache:
            cache[key] = Agent(
                model=load_model(),
                session_manager=get_memory_session_manager(session_id, user_id),
                system_prompt="""
                    You are a personalized tech trend advisor. You remember
                    the user's technical background, skills, and interests
                    from previous conversations. When asked about trends or
                    recommendations, search the web for the latest information
                    and tailor your advice based on what you know about the
                    user. Always be specific and actionable.
                """,
                tools=tools
            )
        return cache[key]
    return get_or_create_agent
get_or_create_agent = agent_factory()
 
 
@app.entrypoint
async def invoke(payload, context):
    log.info("Invoking Agent.....")
 
    session_id = getattr(context, 'session_id', 'default-session')
    user_id = getattr(context, 'user_id', 'default-user')
    agent = get_or_create_agent(session_id, user_id)
 
    stream = agent.stream_async(payload.get("prompt"))
 
    async for event in stream:
        if "data" in event and isinstance(event["data"], str):
            yield event["data"]
 
 
if __name__ == "__main__":
    app.run()

Deploy

ACCOUNT_ID=$(aws sts get-caller-identity --query Account --output text)
cat > agentcore/aws-targets.json << EOF
[{"name":"default","account":"${ACCOUNT_ID}","region":"us-east-1"}]
EOF
 
agentcore deploy -y

Verify Deployment

agentcore status --json

{
  "success": true,
  "projectName": "AgentCoreFull",
  "targetName": "default",
  "targetRegion": "us-east-1",
  "resources": [
    {
      "resourceType": "agent",
      "name": "TechAdvisor",
      "deploymentState": "deployed",
      "detail": "READY"
    },
    {
      "resourceType": "memory",
      "name": "TechAdvisorMemory",
      "deploymentState": "deployed",
      "detail": "SEMANTIC, USER_PREFERENCE, SUMMARIZATION"
    },
    {
      "resourceType": "gateway",
      "name": "my-gateway",
      "deploymentState": "local-only",
      "detail": "1 target"
    },
    {
      "resourceType": "evaluator",
      "name": "ResponseQuality",
      "deploymentState": "deployed",
      "detail": "SESSION — LLM-as-a-Judge — ACTIVE"
    }
  ]
}

All four resources deployed successfully.

Verification: Memory + Gateway Integration

Session 1: Self-Introduction

agentcore invoke \
  "Hi! I'm a platform engineer specializing in Kubernetes and Go. I work on building internal developer platforms. I'm particularly interested in GitOps, service mesh, and observability." \
  --stream \
  --session-id session-full-test-001-self-introduction \
  --user-id user-engineer \
  --agent TechAdvisor

Hello! It's great to meet you! I'll remember that you're a platform engineer
with expertise in:
 
- **Core Skills**: Kubernetes and Go
- **Focus Area**: Building internal developer platforms
- **Key Interests**: GitOps, service mesh, and observability

Session 2: Personalized Trend Search

Wait for Memory's long-term extraction (1–2 minutes, as confirmed in Part 2).

sleep 120

Ask for trends in a new session.

agentcore invoke \
  "What are the latest tech trends that would be relevant to my work? Search the web for recent developments." \
  --stream \
  --session-id session-full-test-002-trend-search-conv \
  --user-id user-engineer \
  --agent TechAdvisor

I'll search for the latest developments in areas most relevant to your
platform engineering work, focusing on GitOps, service mesh, observability,
Kubernetes, and internal developer platforms.

The agent retrieved the user's expertise from Memory and searched the web via Gateway. Here's an excerpt of the personalized response.

## 🕸️ Service Mesh: eBPF Revolution & Sidecarless Architecture
 
**Cilium has won the service mesh performance battle** - adopted as default
CNI by GKE, EKS, and AKS. 40-60% lower latency, 50-70% less memory vs
traditional sidecars.
 
## 📊 Observability: OpenTelemetry Dominance & AI-Powered Insights
 
**71% of orgs use both Prometheus + OTel** (up 50% YoY). New: OTel profiling
(continuous profiling now in OTel spec).
 
## 🎯 Top Actionable Recommendations for You:
 
1. **Pilot Cilium** on a dev cluster - gain eBPF expertise (rare skill)
2. **Build your first golden path** in Backstage with auto-OTel instrumentation
3. **Evaluate Istio Ambient** for service mesh without sidecar overhead

The agent remembered the user's expertise from Session 1 (Kubernetes, Go, GitOps, service mesh, observability), searched the web via Gateway, and returned personalized trend information. Memory and Gateway working together.

Verification: Quality Measurement with Evaluations

After waiting ~10 minutes for trace indexing:

agentcore run evals \
  --agent TechAdvisor \
  --evaluator ResponseQuality \
  --days 1

Agent: TechAdvisor | Mar 23, 2026, 11:35 AM | Sessions: 2 | Lookback: 1d
 
  ResponseQuality: 5.00
 
Results saved to: agentcore/.cli/eval-results/eval_2026-03-23_11-35-41.json

Both sessions scored 5.00 (Excellent). The evaluation prompt includes "personalization based on user context", so Memory-driven personalization is part of the quality assessment.

Summary

• Full integration follows the standard CLI workflow — --no-agent → Gateway → target → agent (--memory longAndShortTerm) → evaluator, then deploy. The CLI auto-generates all integration code. The only manual step was customizing the system_prompt.
• Memory + Gateway is a powerful combination — Remembering user expertise and searching the web for real-time information creates a "knows me and stays current" assistant, built entirely with CLI commands.
• Evaluations can measure integrated agent quality — Including personalization criteria in the evaluation prompt lets you assess how well the agent uses Memory context.
• Region constraints apply — Evaluator CloudFormation support is limited to certain regions. Use us-east-1 or check availability when combining these features.

This series demonstrated that AgentCore CLI's declarative design — centered on agentcore.json and mcp.json — enables consistent agent development from basic deployment through quality management. Understanding each feature individually makes the integration straightforward.

Cleanup

# Remove all resource definitions
agentcore remove all --force
 
# Delete AWS resources
agentcore deploy -y
 
# Uninstall CLI (if no longer needed)
npm uninstall -g @aws/agentcore