1. Read PROMPT.md
2. Answer the Design Questions (write your answers directly in PROMPT.md)
3. Build the prototype:
   • Challenge Track: Build from scratch using PROMPT.md as your spec
   • Guided Track: Go to scaffold/, fill in the TODOs
4. Verify with the MCP inspector tests at the bottom of PROMPT.md
5. Bring your Design Questions answers to live session for discussion

Challenge Track — You decide the architecture, file structure, and implementation. Any language with an MCP SDK works (Python + the official mcp SDK recommended). Read PROMPT.md to get started.

Guided Track — File structure and boilerplate are provided. Fill in the core logic marked with TODO. Go to scaffold/ and follow the instructions below.

cd scaffold
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

You also need Node.js for npx (used by the MCP inspector for verification).

The prototype is a real MCP stdio server. Verify with the MCP inspector (no Claude needed):

npx @modelcontextprotocol/inspector python -m app.mcp_server

This opens a browser GUI — see PROMPT.md Verification section for the full test flow. Once the inspector tests pass, you can optionally connect to Claude Desktop / Claude Code (instructions also in PROMPT.md).

• Connect a real LLM to parse natural language task descriptions before calling task.create
• Add recurring job support (cron expressions)
• Add job chaining (Job A completes -> triggers Job B)
• Add MCP resources support (e.g., expose job details as readable resources)
• Add MCP prompts support (e.g., a daily_review prompt template)