Inspiration

A routine is more than a checklist: people share rooms, tasks depend on earlier steps, and an interruption can invalidate the original plan. Morning Relay explores a practical way to recover while preserving what is already done.

What it does

Morning Relay coordinates people, shared spaces, and task dependencies. Pausing a task explains which later tasks are blocked. Changing an estimate produces a revised plan and an honest late warning. Progress survives a server restart, while revision checks reject stale updates from another view.

How it was built

The local application exposes get_routine, create_routine, and update_task through a real MCP server using protocol 2025-11-25 over Streamable HTTP. A browser board calls those same tools. TypeScript, the official MCP SDK, Express, and Zod provide the implementation; local file persistence preserves routine state.

Challenges and validation

An initial append-only scheduler left an earlier shared-kitchen slot unused. Gap filling corrected the example from 23 minutes to 22, with a focused regression test. Validation includes eight tests, 120 generated dependency schedules, two official SDK clients, a real process restart, and 21 browser MCP requests with no page errors. The demo shows an 18-minute routine, a pause that blocks dependent tasks, persistent completion, a revised 22-minute plan against a 20-minute budget, and stale-update rejection.

What comes next

The prototype currently supports one local routine and one writer process. Estimates and the time budget are entered manually; scheduling is a heuristic rather than a proven optimum. A later step would validate the tool contract with a live Alexa+ host before claiming an Alexa integration.

AI disclosure and scope

The source, design, original fictional example, tests, and English-captioned demo were created with AI assistance. This entry uses the Alexa+ self-hosted MCP route. No live Alexa+, speech, language-model capability, device integration, or certification is claimed. The MIT-licensed source and setup instructions are public. No AWS mini challenge is claimed.

Built With

Share this project:

Updates

Submission history