OmniCare Autonomous

System Architecture – MCP Server ↔ Amazon Bedrock ↔ Ring, Bee & Fire TV
OmniCare Autonomous – Proactive multi-agent caretaking for Alexa+, Ring, Bee & Fire TV
MCP Tools – Five caretaking tools exposed to Alexa+ via Streamable HTTP
Fire TV Alert Dashboard – Multi-modal picture-in-picture emergency UI
Alexa+ Web Simulator – Real MCP client that talks to the self-hosted server

Inspiration

Families with elderly parents or young children constantly worry: “Is Mom okay right now?” Existing smart-home devices are reactive. We wanted a system that is proactive — one that watches biometrics, cameras and ambient context and acts before a crisis escalates.

What it does

OmniCare Autonomous is a multi-modal emergency response + wellness system:

Self-hosted MCP server (Alexa+ track) that exposes caretaking tools over Streamable HTTP (MCP Spec 2025-11-25)
Continuous monitoring via Bee wearable data + Ring cameras
Instant multi-modal alerts pushed to Fire TV as a picture-in-picture dashboard
Multi-agent reasoning powered by Amazon Bedrock (Claude) using AgentCore / Strands-style orchestration
Automated risk mitigation (notify family, log events, escalate)

How we built it

Core: Python MCP server implementing Streamable HTTP
AI layer: Amazon Bedrock (Claude) + agent orchestration loop
Integrations: Ring simulator + Bee data ingestion layer
Front-end: Lightweight Fire TV / web dashboard for live alerts
Fully documented AWS Builder Mini-Challenge integrations

Challenges we faced

No access to gated Alexa+ developer tools → built a fully compliant self-hosted MCP server + web simulator
Time constraints forced prioritisation of a clean, runnable architecture
Balancing real-device requirements with simulator fallbacks so judges can run everything immediately

What we learned

Deep understanding of the Model Context Protocol, how agentic workflows should expose tools safely, and the real friction points when building across Amazon’s device ecosystem.