Inspiration

I'm a student, and I juggle hackathons, assignments and applications at the same time. Every to-do app I tried could show me a list, but none told me the one thing I should start right now. A voice assistant is the perfect place for that, because when you ask "what should I do?", you want a decision, not a list to scroll.

What it does

Deadline Guardian is a self-hosted MCP server for Alexa+ that manages deadlines by voice. It exposes four tools:

  • add_deadline adds a deadline with a due date, effort in hours and priority
  • list_deadlines reads out pending deadlines, soonest first
  • what_should_i_do_now picks the single most at-risk task and warns you if you need more time than you have left
  • complete_deadline marks a task done by its title

Every response is a short, speakable sentence, so it sounds natural when read aloud.

How we built it

  • Node.js and Express
  • Official MCP TypeScript SDK (McpServer)
  • Streamable HTTP transport at /mcp (stateless)
  • Zod schemas for tool inputs
  • A local JSON file for persistence
  • Tested end to end with the MCP Inspector

The "what should I do now" tool ranks tasks by a simple risk score:

$$ \text{risk} = \frac{\text{effort hours}}{\text{hours left}} \times \text{priority weight} $$

The higher the score, the sooner you should start.

Challenges we ran into

  • Building the whole project in the final hours of the hackathon, so I kept the scope small and the demo reliable.
  • Writing tool responses for the ear instead of the eye. Short sentences and spoken durations ("3 days left") work better than timestamps.
  • Making the date input forgiving, so a bad date gets a friendly spoken error instead of a crash.

Accomplishments that we're proud of

  • A fully working MCP server with four tools, built and tested within the hackathon window
  • A voice-first design where every answer is a clear decision, not a data dump
  • A risk-based ranking that warns you when a task can't fit in the time left
  • A clean, open-source repo that anyone can run with npm install and node server.js

What we learned

MCP makes it easy to turn plain code into something a voice assistant can use. Designing outputs for voice changes how you think about an interface: shorter, clearer, and decision-first. I also learned how stateless Streamable HTTP keeps a server simple to build and test.

What's next for Deadline Guardian

  • Account linking and per-user storage
  • Proactive spoken reminders through Alexa+
  • Calendar sync with Google Calendar
  • Smarter effort estimates that learn from how long your tasks actually take

Built With

Share this project:

Updates

Submission history