Curate intentional capsules with persistent Val Town templates, OpenAI-completion friendly tools, and an interactive shopping checklist widget.

• Backend: Hono + MCP Lite + TypeScript capsule templates
• Widget: React 19 + TanStack Router + OpenAI Apps SDK hydration
• Shared Types: Zod schemas for structured content and widget state

1. Visit your deployment URL to copy the /mcp endpoint.
2. In ChatGPT, enable developer mode and add the MCP server with that endpoint.
3. Try prompts such as:
   • “Create a professional women capsule for $850 in the fall.”
   • “Finalize the capsule shopping list with extra budget for shoes.”

• capsule_start({ style, budget, gender, season })
  Seeds a capsule configuration using persistent templates and scales category budgets.
• capsule_generate_list({ capsule_id, preferences })
  Generates the full shopping list with category tracking, priority groupings, and widget hydration data.

Both tools return discriminated unions that drive the widget:

Create val
type CapsuleStructuredContent =
  | { kind: "capsule_overview"; capsule: CapsulePlan }
  | { kind: "capsule_shopping_list"; capsule: CapsulePlan };

CapsulePlan includes category allocations, priority-tagged items, totals, and preference metadata so the widget can render progress bars, badges, and persistent checkbox state.

• Overview screen with budget progress, template notes, and preference highlights.
• Shopping checklist with persistent checkboxes, live budget tallies, and export-to-txt.
• Mobile-friendly layout with CSS progress bars and priority badges.

1. Extend backend/mcp/capsuleTemplates.ts with additional personas or seasons.
2. Enrich capsule_generate_list with smarter preference adjustments (color palettes, exclusions, etc.).
3. Update the widget styling or add alternative export formats (CSV, Notion, Airtable).