CareLoop CareLoop is an Alexa+ style family-care coordinator. It gives caregivers one place to see check-ins, manage reminders, and ask an AI assistant to take action.

Hackathon entry plan: Alexa+ primary track with the AWS Builder mini challenge. See the hackathon compliance plan for the requirements checklist and demo plan.

What is included A responsive caregiver dashboard with synthetic family data A Node.js and Express API An assistant chat interface A tool registry shared by the AI agent and MCP server A self-hosted MCP endpoint at POST /mcp using Streamable HTTP Local demo mode that works without AWS credentials Optional Amazon Bedrock tool use through the Converse API Reminder escalation and caregiver notification simulation Run locally Requirements: Node.js 22 or newer LTS.

Bash

cd careloop npm install Create the environment file:

macOS/Linux:

Bash

cp .env.example .env Windows CMD:

cmd

copy .env.example .env Start the application:

Bash

npm run dev Open http://localhost:3000.

The local demo uses BEDROCK_ENABLED=false and synthetic family data. This lets the product work while AWS account verification or Bedrock access is pending.

Useful commands Bash

npm run typecheck npm run dev API routes GET /health GET /api/overview GET /api/members GET /api/tasks GET /api/activity POST /api/chat with { "message": "Is Dad okay?" } POST /api/escalations/check — local stand-in for the future scheduled Lambda worker POST /api/escalations/check?force=true — demo-only fast-forward for testing an escalation POST /mcp — Streamable HTTP MCP endpoint MCP tools get_status create_reminder notify_caregiver record_checkin get_daily_summary Demo prompts Try these in the assistant:

Is Dad okay? Give me a summary of everyone today Remind Dad about his medication Notify Sarah about Dad Dad completed his check-in Dad hasn't checked in. Remind him in 20 minutes and notify Sarah if he doesn't respond. The combined request creates a reminder with an escalation policy. For a fast demo test, call POST /api/escalations/check?force=true instead of waiting for the scheduled due time. In production, this endpoint will be replaced by a scheduled AWS Lambda worker.

Bedrock configuration Set these values in .env after AWS access is available:

env

BEDROCK_ENABLED=true AWS_REGION=us-east-1 BEDROCK_MODEL_ID=amazon.nova-lite-v1:0 MCP_SERVER_URL=http://127.0.0.1:3000/mcp When Bedrock is enabled, the intended flow is:

text

User → Bedrock → tool selection → MCP → CareLoop action → updated dashboard If Bedrock is disabled or unavailable, CareLoop uses the local fallback mode.

AWS and production phases Validate the product flow with local synthetic data. Enable Amazon Bedrock and document the Converse API integration. Replace the in-memory store with DynamoDB behind the same tool interface. Replace the local escalation timer with a scheduled AWS Lambda worker. Deploy the API and expose a secure HTTPS MCP URL for the Alexa+ integration path. Record the final demo and submit the project on Devpost. Hackathon submission checklist Choose Alexa+ as the primary track. Choose AWS Builder as the mini challenge after the Bedrock integration is verified. Demonstrate the simulated Alexa+ conversation in the video. Show the MCP server and real tool calls in the repository and demo. Include the public GitHub repository and MIT license. Record a public English demo video under three minutes. Include product feedback and a friction log for every tool, API, or SDK used. Never commit AWS access keys or real health information. Use synthetic data for the hackathon demo.