Build a Zod .input() Validator for tRPC Procedures from Excel Data

Single-record mutation input

Generate from one row, name it importInput, and use it directly in .input().

Sample row -> JSON: { "name": "Sue", "email": "sue@x.com", "seats": 5 }
rootName: importInput

Generated:
export const importInput = z.object({
  name: z.string(),
  email: z.string(),
  seats: z.number().int()
});

Router:
addMember: publicProcedure
  .input(importInput)
  .mutation(({ input }) => db.member.create({ data: input }))

Batch import with z.array()

Wrap the row schema in z.array() so the whole uploaded sheet is validated as one mutation input.

import { importInput } from "./schema";

importRows: publicProcedure
  .input(z.array(importInput))
  .mutation(async ({ input }) => {
    // input is typed importInput[]; any bad row already rejected
    await db.member.createMany({ data: input });
    return { inserted: input.length };
  })

Add .email() and .optional() the generator won't

Structure comes from the generator; the semantic refinements are yours.

Generated: email: z.string(), nickname: z.string(),

Refined for the procedure:
email: z.string().email(),
nickname: z.string().optional(),

Now .input() rejects bad emails and accepts a missing nickname.

Strict input to reject smuggled fields

strictObjects makes an unexpected payload key a validation error before the resolver runs.

strictObjects: true ->
export const importInput = z.strictObject({
  name: z.string(),
  seats: z.number().int()
});

Client sends { name, seats, role: "admin" }:
tRPC returns a BAD_REQUEST input error, 'Unrecognized key: role'
role never reaches the resolver.

Full Excel-native router in one pass

When you want schema + interface + CRUD from the sheet itself, use the tRPC Router Builder.

/excel-tools/excel-trpc-router on a Members sheet (routerName: memberRouter):

export const MembersSchema = z.object({
  name: z.string().optional(),
  seats: z.number().optional()
});
export interface Members { name?: string; seats?: number; }

export const memberRouter = router({
  list: publicProcedure.input(z.object({
    page: z.number().int().min(1).default(1),
    pageSize: z.number().int().min(1).max(100).default(20)
  })).query(/* ... */),
  // getById, create, update, delete also generated
});

The redirect target takes JSON only. Convert one row with sheet_to_json first, or use the tRPC Router Builder, which reads .xlsx/.csv directly.

The JSON to Zod generator outputs only the schema and a z.infer<> alias — you compose .input(...) yourself. The tRPC Router Builder is the one that emits full .input()/procedure scaffolding.

The JSON to Zod generator never adds .optional(), so a non-required field present in your sample becomes required and rejects payloads that omit it. Add .optional(), or use the tRPC Router Builder, which makes every field optional.

Dates infer z.string(). tRPC serialises Dates over the wire; for inputs use z.coerce.date() (or z.string().datetime()) so the resolver receives the type it expects.

z.array(rowSchema) validates the entire array as the procedure input — one bad row fails the whole mutation with a BAD_REQUEST. For partial-success imports, accept z.array(z.unknown()) and safeParse each row inside the resolver instead.

Default z.object() strips unknown keys, so an unexpected field passes and disappears from input. Use strictObjects: true (-> z.strictObject()) to reject it as a BAD_REQUEST.

The generated schema is standard Zod, so it works identically as .input() in tRPC v10 and v11. The tRPC Router Builder's emitted procedures use the publicProcedure builder pattern common to both.

The JSON to Zod generator JSON.parses the input; a malformed paste throws and yields nothing. Use JSON.stringify(rows[0], null, 2) for a clean sample.

JSON to Zod is Pro with a 2 MB free cap. One row infers the schema — never paste the entire dataset.

z.number() rejects "5". Use z.coerce.number() in the input schema if your client sends numeric strings, or keep it strict to force the client to send real numbers.