Inspiration
Household task lists rarely resolve who can do each task, how much time is available, or which agreement is current. HearthRelay makes the handoff explicit and preserves it between sessions.
What it does
Capture tasks with time estimates, priority and eligible people. Compute an exact allocation for supplied time budgets, explain deferred tasks, and save a proposal. Confirmation rejects stale state and is idempotent. Export confirmed tasks as calendar events.
How it was built
Python, the official MCP Python SDK, Streamable HTTP and SQLite. A dynamic-programming planner provides exact assignments for up to twelve tasks and three people. Six tools cover capture, state inspection, proposal, confirmation, completion and calendar export.
Challenges and lessons
Proposals must remain valid when confirmed and survive retries without duplicate writes. State-version checks and idempotent confirmation handle those cases. The first local MCP client also inherited a system SOCKS proxy; the loopback verifier now explicitly disables environment proxies.
Verified demonstration
Four core tests and a real HTTP MCP session passed, negotiating protocol 2025-11-25. In the fictional demonstration, four tasks compete for Alex's 25 minutes and Robin's 15 minutes. Dinner, tomorrow's bags and pet water are assigned; laundry is deferred. The client confirms twice, exports three calendar events and completes one task.
Current scope and next steps
This is a local, self-hosted MCP prototype. No actual Alexa device or Amazon-hosted integration has been tested. Authentication and a client-side human approval interface are required before shared deployment. Public source and demo-video publication are in progress.
AI assistance
OpenAI Codex developed the code, tests, documentation and submission text under Dominique Wrase's instructions. All household data is fictional; no real-world impact is claimed.
