HomeCare Copilot: AI Appliance Assistant
Inspiration
Household appliances make everyday life easier, but troubleshooting them can be frustrating. A dishwasher leaving cloudy dishes, a washing machine with confusing settings, or a vacuum approaching the end of its brush life can leave users searching through manuals, guessing at solutions, and struggling to find compatible replacement parts.
I wanted to explore a better approach: what if an AI assistant could do more than answer questions and actually coordinate a complete appliance-care workflow?
That idea inspired HomeCare Copilot, an AI-powered appliance assistant built around the Model Context Protocol (MCP).
What it does
HomeCare Copilot connects conversational requests with structured appliance-care tools. Users can describe problems naturally and receive diagnostic insights, maintenance recommendations, supported device settings, and replacement-part suggestions.
Key features include:
- Appliance diagnosis: Uses simulated device telemetry and diagnostic rules to identify likely issues.
- Transparent recommendations: Presents the leading suspected cause, supporting evidence, confidence information, and alternative possibilities.
- Capability-aware settings: Maps user requirements to supported appliance settings and constraints.
- Controlled actions: Supports dry-run previews before appliance state changes.
- Safer parts ordering: Separates purchase requests from explicit confirmation and applies server-side spending policies.
- Maintenance monitoring: Tracks simulated appliance health and component wear.
- Tool visibility: Makes agent activity and results easier to inspect and understand.
How we built it
The project uses Python, FastAPI, the MCP Python SDK, and Streamable HTTP to connect a simulated Alexa+-style host with an appliance-care MCP server.
The server exposes structured tools for listing appliances, retrieving device state, diagnosing issues, suggesting settings, executing actions, finding parts, requesting purchases, and confirming purchases.
Mock appliance state machines represent a dishwasher, washing machine, and robot vacuum. Deterministic diagnostic logic and server-side policy checks help keep device operations predictable, while the conversational planner translates user requests into appropriate tool workflows.
A browser-based interface provides a place to interact with the assistant and inspect its responses. Automated integration tests help verify communication between the host and MCP server.
MCP was an important architectural choice because it separates the conversational interface from the capabilities it can discover and invoke.
Challenges we faced
One key engineering challenge was connecting the conversational planner to the MCP server while maintaining clear tool contracts and reliable communication.
Another challenge was making the assistant's decisions understandable. Instead of presenting only a diagnosis, we improved the planner to expose the leading suspected cause, confidence information, ranked alternatives, telemetry evidence, and dry-run action parameters.
Safety was another important consideration. An AI-generated recommendation should not automatically authorize a consequential action. HomeCare Copilot therefore uses a separate policy layer for supported operations and a two-step purchase workflow for simulated transactions.
Working within a short development window also required prioritizing the core workflow, integration testing, and demonstrable functionality over adding numerous unfinished features.
What we learned
This project reinforced three important lessons.
First, AI agents become more useful when they can interact with structured tools instead of relying only on generated text.
Second, consequential operations need deterministic safeguards outside the conversational model. Prompts alone are not a substitute for server-side validation, spending limits, and explicit confirmation.
Finally, transparency matters. Showing evidence, tool activity, and proposed actions makes an agent easier to understand, debug, and evaluate.
What's next
Future improvements could include integration with real smart-home platforms, manufacturer-specific diagnostic information, richer multimodal troubleshooting, more extensive evaluation scenarios, and production-grade authorization for real commerce.
The current implementation focuses on simulated appliances and purchases, providing a foundation for exploring these capabilities safely.
Current limitations
HomeCare Copilot is a prototype. Appliance telemetry, device operations, and purchases are simulated. It does not currently provide official Amazon Alexa+ integration, real appliance control, or real payment processing.
The project is designed to demonstrate how MCP can connect conversational AI with structured capabilities while keeping important actions controlled and observable.
Log in or sign up for Devpost to join the conversation.