Write Zod validation schemas from a real API response instead of by hand
Runs in your browser. Nothing you type is sent anywhere. Checked against sources on 22 Oct 2026.Validating an API response with Zod means writing a schema that matches it field by field. Paste one real response and this writes it for you, with one named schema per nested object, children declared before their parents, and a z.infer type for each. Keys that only some array items have get .optional(), null values get .nullable(), and strings that are all emails, URLs, UUIDs or ISO dates get that check.
The page opens with Named selected, so every nested object gets its own schema.
Paste one real JSON response into the JSON pane.
Change Root name to the name of your top-level schema; nested schemas are named from their keys.
Read the schemas in the Zod pane and click Copy to paste them into your code.
This JSON to Zod converter reads a sample of your data and writes the matching Zod schema, so you can validate what an API, a form or a config file really sends at runtime and get the TypeScript type from the same source. The output updates as you type.
Inline or named. Inline writes one schema with everything nested inside it. Named writes one constant per object shape, children before their parents so the file runs top to bottom, reuses one schema for identical shapes, and adds a z.infer type for each.
The code uses calls that work in both Zod 3.23 and later and Zod 4. Every case in the tests is run through the real Zod library to check that the schema accepts the sample it came from. Nothing you paste is sent anywhere. Related tools: JSON to TypeScript, JSON Schema Validator, JSON Formatter.
The rest of the article, and all 6 questions about JSON to Zod, are on the JSON to Zod page.
Paste the JSON into the left pane and the schema appears on the right as you type. Set the root name, choose Inline or Named, then press Copy or Ctrl+Enter and paste the result into a .ts file that imports z from zod.
Yes. The output only uses calls both versions have: z.object, z.array, z.union, z.record with two arguments, .optional(), .nullable(), .nullish(), .strict() and the string checks. Zod 4 marks some string checks as deprecated in favour of z.email() and similar, but they still work.
A format check is added only when every sample of that field matches it. One value that is not an email, or a date-time without seconds or with an offset written +0530, keeps the field a plain z.string(), because Zod would reject that value.
Inline writes a single schema with every nested object written inside it, which is easy to read for small payloads. Named writes one constant per object shape, such as AddressSchema, so you can reuse and export each part, and identical shapes share one schema.
Rates and rules change; check the source before you rely on a figure.
Still stuck? Ask about JSON to Zod and anybody who knows can answer. No account needed.
On the related tools: TSV to JSON – Convert tab-separated values, such as a paste from Excel, into JSON · Query String Parser – Split a URL query string into a table of decoded parameters.