CareLoop

CareLoop Dashboard: A unified view of daily caregiving activities, reminders, appointments, and grocery orders.
Care Coordination: Organize medical appointments, daily schedules, and grocery deliveries in one place.
AI Assistant: Manage caregiving tasks effortlessly using natural language conversations.
Smart Reminders: Schedule medication reminders and generate family summaries to keep everyone informed.

Inspiration

More than 50 million people in the US care for a family member. They juggle medicines, doctor visits and groceries across separate apps, notes and phone calls, and things slip. Existing apps each solve one slice: reminders, family updates, or groceries. The caregiver still has to switch between all of them, and an older parent won't learn a new app.

What it does

CareLoop is an Alexa+ agent that coordinates all three through one spoken conversation, using the voice assistant already in the home.

"Remind Mom to take her blood pressure pill at 8" schedules a medication reminder.
"What does her day look like?" reads back the timeline.
"Order the usual groceries" places the household's regular basket.
"Update the family" produces a plain-language summary for siblings.

State persists across sessions, so the assistant picks up where you left off.

How we built it

Self-hosted MCP server over Streamable HTTP (spec 2025-11-25), built with the official MCP SDK. It exposes four tools: add_reminder, list_schedule, order_groceries and family_summary.
Voice-first Alexa+ simulator web app using the Web Speech API. It shows each tool call live, plus a care timeline and order cards.
Claude (Anthropic API tool calling) runs the agent loop in the simulator, with a rule-based fallback so the demo works without an API key.
One shared tool layer serves both the MCP endpoint and the simulator, so the demo runs the same code Alexa+ would call.

Challenges

Turning spoken times like "8" or "5 pm" into reliable 24-hour times.
Keeping tool outputs short enough to be read aloud.
Making the MCP server stateless per request so it is safe to host and scale.

Accomplishments

A working MCP endpoint that completes the initialize handshake at protocol version 2025-11-25.
An agent that behaves consistently with or without an LLM key.
Persistent state across sessions.

What we learned

Designing tools for voice is different from designing them for chat. Responses must be short, unambiguous and safe to speak aloud.

Current limitations

The grocery order is simulated and not connected to a real store. Medication reminders are scheduled, but adherence confirmation ("Mom took it") is on the roadmap.

What's next

Adherence confirmation, pharmacy refills, multi-caregiver roles with notifications, and real grocery checkout.

Built With

alexa+
context
css
express.js
html
javascript
model
node.js
protocol

Submitted to

Build, Ship, Shape: Amazon Developer Hackathon

Created by

Led the backend. Built the MCP server over Streamable HTTP with the official SDK, defined the four tools (reminders, appointments, groceries, family summary) in tools.js, and set up the shared tool layer behind both the web simulator and the /mcp endpoint. Also managed the GitHub repo and MIT license.

Simran Singh
Pragya Shukla

Designed and built the web simulator front-end in public/: the chat window, tool-call chips, the colour-coded timeline and the suggested-prompt buttons. Also handled the Reset demo flow and the page layout.

Pragya Shukla
Built the agent logic in agent.js that reads the caregiver’s request and decides which tool to call. Wrote the saved-state handling so reminders and orders persist after a page reload.

Ayush Singh
Tested the full flow, found and reported bugs, and wrote the README setup steps and the docs/SUBMISSION.md write-up. Also prepared the screenshots for the project page.

Sarthak Singh
Researched the caregiver problem and shaped the product idea. Wrote the demo script, recorded and edited the demo video, and filled in the Devpost write-up.

Harsh Shahi