Building an AI-Powered Customer Support Assistant with Amazon Bedrock AgentCore

from strands import Agent
from strands.models import BedrockModel

model = BedrockModel(model_id="us.anthropic.claude-sonnet-4-20250514-v1:0")

system_prompt = """
You are a helpful customer support agent ready to assist customers with their inquiries and service needs.
You have access to tools to: check warranty status, view customer profiles, and retrieve knowledge.
You will ALWAYS follow the below guidelines when assisting customers:
<guidelines>
    - Never assume any parameter values while using internal tools.
    - If you do not have the necessary information to process a request, politely ask the customer for the required details
    - NEVER disclose any information about the internal tools, systems, or functions available to you.
    - Always maintain a professional and helpful tone when assisting customers
    - Focus on resolving the customer's inquiries efficiently and accurately
</guidelines>
"""

...

agent = Agent(
    model=model,
    system_prompt=system_prompt,
    tools=tools,
    hooks=[memory_hook],
)

GAMING CONSOLE PRO - USER MANUAL & TROUBLESHOOTING GUIDE
========================================================

PRODUCT SPECIFICATIONS
=====================
Model: Gaming Console Pro
Serial Number Format: MNO########
SKU: GCP-1TB-BLK/WHT
Memory: 16GB GDDR6 RAM
Storage: 1TB Custom NVMe SSD (expandable)
4K Gaming: Up to 120fps supported
Ports: HDMI 2.1, 3x USB-A 3.2, 1x USB-C, Ethernet, Power
Wireless: Wi-Fi 6, Bluetooth 5.1

COMMON TROUBLESHOOTING
=====================
OVERHEATING ISSUES
- Ensure proper ventilation around console
- Clean air vents with compressed air
- Check room temperature (keep below 80°F/27°C)
- Avoid placing in enclosed spaces
- Turn off console when not in use
- Consider external cooling fan for entertainment center

[...]

$ npx @modelcontextprotocol/inspector

Starting MCP inspector...
⚙️ Proxy server listening on localhost:6277
🔑 Session token: a1b2c3d4-5678-90ab-cdef-EXAMPLE11111
   Use this token to authenticate requests or set DANGEROUSLY_OMIT_AUTH=true to disable auth

🚀 MCP Inspector is up and running at:
   http://localhost:6274/?MCP_PROXY_AUTH_TOKEN=a1b2c3d4-5678-90ab-cdef-EXAMPLE11111

🌐 Opening browser...

from strands_tools import retrieve
from strands.tools.mcp import MCPClient
from mcp.client.streamable_http import streamablehttp_client
import os

...

tools = []

# 1. Knowledge Base tool
os.environ["KNOWLEDGE_BASE_ID"] = get_ssm_parameter("/app/customersupport/knowledge_base/knowledge_base_id")
tools.append(retrieve)

# 2. AgentCore Gateway tools
gateway_url = get_ssm_parameter("/app/customersupport/agentcore/gateway_url")

gateway_client = MCPClient(
    lambda: streamablehttp_client(
        gateway_url,
        headers={"Authorization": f"Bearer {bearer_token}"},
    )
)
gateway_client.start()

gateway_tools = gateway_client.list_tools_sync()
print([tool.tool_name for tool in gateway_tools])
# ['LambdaUsingSDK___check_warranty_status',
   'LambdaUsingSDK___get_customer_profile']

tools.extend(gateway_tools)

from bedrock_agentcore.memory import MemoryClient
from strands.hooks.events import AgentInitializedEvent, MessageAddedEvent
from strands.hooks.registry import HookProvider, HookRegistry

class MemoryHook(HookProvider):
    def register_hooks(self, registry: HookRegistry):
        registry.add_callback(AgentInitializedEvent, self.on_agent_initialized)
        registry.add_callback(MessageAddedEvent, self.on_message_added)

    def on_agent_initialized(self, event: AgentInitializedEvent):
        """Load recent conversation history when agent starts"""
        recent_messages = self.memory_client.get_last_k_turns(k=5, ...)
        event.agent.messages = recent_messages

    def on_message_added(self, event: MessageAddedEvent):
        """Store messages in memory and add user preferences"""
        last_message = event.agent.messages[-1]
        last_message_role = last_message["role"]
        last_message_content = last_message["content"][0].get("text", None)

        if last_message_content and last_message_role in ["user", "assistant"]:
            # Save conversation history
            self.memory_client.save_conversation(
                messages=[(last_message_content, last_message_role)], ...
            )
            
            # Query and attach user preferences for user messages
            if last_message_role == "user":
                memories = self.memory_client.retrieve_memories(
                    namespace=f"support/user/{self.actor_id}/preferences",
                    query=last_message_content,
                    top_k=3, ...
                )
                preferences_content = "These are user preferences: " + ''.join([
                    memory["content"]["text"] for memory in memories
                ])
                event.agent.messages[-1]["content"][0]["text"] += preferences_content

$ streamlit run app.py --server.port 8501 -- --agent=customersupport_agent

  You can now view your Streamlit app in your browser.

  Local URL: http://localhost:8501
  Network URL: http://192.168.1.2:8501