runtime

AgentCore Runtime Deployment

Deploy a Valyu-powered research agent to AWS Bedrock AgentCore Runtime.

Architecture

User Request
     │
     ▼
┌─────────────────────────────┐
│   AgentCore Runtime         │
│   (Serverless container)    │
│                             │
│   ┌─────────────────────┐   │
│   │   Strands Agent     │   │
│   └──────────┬──────────┘   │
│              │              │
└──────────────┼──────────────┘
               │ MCP + OAuth
               ▼
┌─────────────────────────────┐
│   AgentCore Gateway         │
│   (Centralized tool access) │
└──────────────┬──────────────┘
               │
               ▼
        ┌──────────────┐
        │  Valyu MCP   │
        └──────────────┘

Prerequisites

1. AWS CLI configured (aws configure)
2. Python 3.10+
3. Valyu API key from platform.valyu.ai

Complete Setup (Start to Finish)

Step 1: Install dependencies

pip install "valyu-agentcore[agentcore]"

Step 2: Set up Gateway with Valyu

First, create a Gateway with Valyu tools:

cd /path/to/valyu-agentcore/examples/gateway

export VALYU_API_KEY=your-valyu-api-key

python setup_gateway.py

This creates valyu_gateway_config.json in the gateway folder.

Step 3: Copy config to runtime folder

cp valyu_gateway_config.json ../runtime/
cd ../runtime

Step 4: Configure and deploy the agent

# Configure (creates .bedrock_agentcore.yaml)
agentcore configure --entrypoint agent.py --non-interactive --name valyuagent

# Deploy to AWS (uses CodeBuild, no Docker needed)
agentcore launch

# Test it
agentcore invoke '{"prompt": "What is the current stock price of NVIDIA?"}'

That's it!

The agent is now running in AWS and can answer questions using Valyu search tools.

Alternative: Environment Variables

Instead of copying the config file, you can set environment variables when deploying:

agentcore launch \
  --env GATEWAY_URL=https://your-gateway.gateway.bedrock-agentcore.us-east-1.amazonaws.com/mcp \
  --env COGNITO_CLIENT_ID=your-client-id \
  --env COGNITO_CLIENT_SECRET=your-client-secret \
  --env COGNITO_DOMAIN=your-cognito-domain \
  --env COGNITO_SCOPE=your-gateway-name/invoke

How It Works

1. Agent starts in Runtime container
2. Reads Cognito credentials from config file or environment
3. Gets fresh OAuth token from Cognito on each request
4. Calls Gateway with Bearer token
5. Gateway routes request to Valyu MCP
6. Valyu returns search results
7. Agent uses results to answer the question

Files

File	Purpose
agent.py	The agent code that runs in Runtime
requirements.txt	Python dependencies
valyu_gateway_config.json	Gateway credentials (you copy this from gateway folder)

Troubleshooting

"No such file: valyu_gateway_config.json"

• You need to copy the config from the gateway folder
• Run: cp ../gateway/valyu_gateway_config.json .

"GATEWAY_URL not found"

• Either copy the config file OR set environment variables
• See "Alternative: Environment Variables" above

Token/auth errors

• Make sure you ran setup_gateway.py first
• Check that the config file has valid credentials

"Agent not found"

• Run agentcore configure before agentcore launch

Cleanup

To delete the deployed agent:

# List agents
aws bedrock-agentcore-control list-agent-runtimes --region us-east-1

# Delete agent
aws bedrock-agentcore-control delete-agent-runtime \
  --agent-runtime-id YOUR_AGENT_ID \
  --region us-east-1

To delete the Gateway (from the gateway folder):

cd ../gateway
python setup_gateway.py cleanup