Turn an Excel Requirements Sheet into a Working tRPC API Prototype

Events entity to a list-and-detail prototype

An Events sheet generates a router whose list (paginated + searchable) and getById back a table-plus-detail UI immediately. Stub ctx.db with an array to demo before the DB exists.

Input (sheet: Events):
id,title,startDate,capacity,published
1,Launch Party,2026-07-01,200,true
2,Webinar,2026-07-15,500,false

Router name: eventsRouter

Generated eventsRouter.ts gives you:
  list:    publicProcedure (page/pageSize/search) -> ctx.db.event.findMany
  getById: publicProcedure ({ id: uuid }) -> ctx.db.event.findUnique
  create/update/delete: protectedProcedure

Inferred types:
  id z.number(), title z.string(), startDate z.string() (date->string),
  capacity z.number(), published z.boolean()

Demo with an in-memory stub before the DB

For a quick prototype, replace the Prisma placeholder with an array so the endpoints work without any database wiring.

// Generated: const data = await ctx.db.event.findMany({ ... });

// Quick prototype stub:
const MEM: Event[] = [{ id: 1, title: "Launch Party", capacity: 200, published: true }];
list: publicProcedure
  .input(/* same generated input */)
  .query(({ input }) => {
    const start = (input.page - 1) * input.pageSize;
    return { data: MEM.slice(start, start + input.pageSize), page: input.page, pageSize: input.pageSize };
  }),

Searchable columns are the first two strings

Search in the prototype table works on the first two z.string() columns automatically. Order your headers so the columns you want to search come first.

Headers order matters for search:
id, title, location, description
  title    -> 1st string (searched)
  location -> 2nd string (searched)
  description -> 3rd string (NOT searched by default)

To search description too, add it to the generated OR array:
  { description: { contains: search, mode: "insensitive" } }

Iterate: add a column, regenerate, diff

Add an organizer column to the sheet and regenerate. Deterministic output means a clean diff shows exactly the new field.

v1 headers: id, title, capacity
v2 headers: id, title, capacity, organizer

Diff of generated schema:
  ProductSchema = z.object({
    id: z.number().optional(),
    title: z.string().optional(),
    capacity: z.number().optional(),
+   organizer: z.string().optional(),
  });

// Same input -> same output, so version control shows just the delta.

Multi-screen prototype from one workbook

Tabs for Users, Events, and Tickets generate a root router with three sub-routers in a single pass — enough to back a three-screen prototype.

Workbook tabs: Users | Events | Tickets
Router name: apiRouter

export const apiRouter = router({
  users:   usersRouter,
  events:  eventsRouter,
  tickets: ticketsRouter,
});
export type ApiRouter = typeof apiRouter;
// Mount apiRouter on your root appRouter and all three screens have endpoints.

Prototyping on a Free / Pro / Pro+Media account fails — the tool throws tRPC Router Builder requires Developer tier. and shows an upgrade overlay. Use a Developer-tier account (or runner key) for the generate.

ISO YYYY-MM-DD columns are typed z.string(), not z.date(). For a prototype this is usually fine, but if a form binds to a date picker you may want z.coerce.date() — change those fields after generating.

The base schema makes everything .optional(), so reads accept partial records. create re-requires all fields via .required(). For a quick prototype you may want to relax create for fields you haven't built inputs for yet.

Only the first two z.string() columns are wired into search. If your demo's search box returns nothing for a third text field, add it to the generated OR array manually.

getById expects z.string().uuid() but a spreadsheet id is usually numeric. In a stub-backed prototype, either pass UUID strings as ids or change the getById/delete input to z.number() to match your in-memory data.

SheetJS uses row 1 as field names. A throwaway sheet that starts with data will produce fields named after those first values. Keep a header row at the top even for scratch sheets.

In multi-tab mode the entity name comes from the sheet name. Leaving tabs as Sheet1 / Sheet2 produces Sheet1 / Sheet2 entities. Rename tabs to the entities you're prototyping.

A column mixing numbers and the odd text value (42, TBD) can't be all-numeric, so it falls back to z.string(). Clean throwaway data, or accept the looser type for the prototype.

Large multi-entity outputs are cut off in the on-page preview. The complete file is available via Copy (full string) or Download — don't paste from a truncated preview.

A sheet with headers but zero data rows still works (types fall back to string), but a workbook with no data rows at all returns // No data. Add a sample row so inference has something to read.