How is this different from the Zod Schema Generator tool?

The Zod Schema Generator turns TypeScript types or JSON examples into Zod. This tool is the JSON Schema bridge: it converts a real JSON Schema document into Zod and — uniquely — converts Zod source back into JSON Schema. If you are working with OpenAPI specs, function-calling tool definitions, or structured-output schemas, this is the converter you want.

What is "strict mode" and why does it matter for LLMs?

When generating JSON Schema from Zod, strict mode recursively adds `additionalProperties: false` to every object and lists every property in `required`. OpenAI Structured Outputs and Anthropic tool use require this exact shape — no extra keys, all fields present — so the model is forced to return a value that matches your schema exactly. One checkbox produces a schema you can drop straight into a tool definition.

Does the reverse direction run or eval my Zod code?

No. The Zod → JSON Schema direction uses a small hand-written tokenizer and recursive-descent parser that reads a documented subset of Zod source as text. Nothing is executed, so it is safe to paste schemas you do not fully trust, and it works entirely offline in your browser.

Which Zod features does the reverse parser understand?

It supports z.object, z.string, z.number, z.boolean, z.null, z.array, z.enum, z.literal, and z.union, plus the chained modifiers .optional(), .nullable(), .min(n), .max(n), .int(), .email(), .url(), .uuid(), .datetime(), .regex(/…/), .default(x), .describe('…') and .strict(). Anything outside that subset is skipped with a best-effort result and an inline // note so you can see what was dropped.

How are JSON Schema $defs and $ref handled?

Local references like #/$defs/Address or #/definitions/Address are resolved inline — the referenced sub-schema is expanded in place wherever it is used, producing self-contained Zod with no external lookups. References that cannot be resolved fall back to z.unknown() and are flagged with a note.

What happens with input the tool cannot fully convert?

It never throws. Invalid JSON, an unexpected root, an unresolved $ref, or an unsupported Zod construct all produce a usable result whose first line is a // note: comment describing exactly what could not be translated, plus a safe placeholder (z.unknown() or an empty schema) so the rest of your output stays intact.

JSON Schema ⇄ Zod Converter

Convert JSON Schema to Zod and Zod back to JSON Schema — built for LLM structured outputs and function calling. Maps formats, enums, const, unions, $ref, nullable, defaults and descriptions, with a one-click strict mode (additionalProperties:false + all required) for OpenAI/Anthropic tool definitions. Runs entirely in your browser.

About this ToolHow it works, benefits & use cases

The JSON Schema ⇄ Zod Converter is a two-way bridge between JSON Schema (draft 2020-12 / OpenAPI-style) and Zod schema source, built specifically for the LLM era where the same shape has to live in two places at once: a JSON Schema for an OpenAI or Anthropic function-calling / structured-output request, and a Zod schema for validating the model’s reply in TypeScript. Going forward, paste a JSON Schema and get clean Zod source — objects with proper required vs `.optional()` fields, `additionalProperties:false` rendered as `.strict()`, string formats mapped to `.email()` / `.url()` / `.uuid()` / `.datetime()`, numeric and length bounds, `enum` and `const`, `anyOf`/`oneOf` unions, `allOf` intersections, nullable types, defaults, descriptions, and local `$ref` pointers resolved inline. Going in reverse, paste a common subset of Zod and get back pretty-printed JSON Schema you can hand straight to a model’s tool definition, with an optional strict mode that adds `additionalProperties:false` and marks every property required — exactly what the major function-calling APIs demand. Everything runs in your browser, never throws on bad input, and flags anything it can’t fully translate with an inline note so you’re never guessing.

How to Use

1Choose a direction at the top: "JSON Schema → Zod" to generate Zod source, or "Zod → JSON Schema" to generate a schema.
2Paste your source into the editor on the left (JSON Schema, or a z.object({…}) expression), or hit a "Load example" button.
3Set the schema / const name used in the generated Zod export, e.g. User or SearchArgs.
4For JSON Schema → Zod, tick "Export inferred type" to also emit `export type Name = z.infer<typeof Name>`.
5For Zod → JSON Schema, tick "Strict mode" to add additionalProperties:false and require every property — what OpenAI/Anthropic function-calling needs.
6Copy, download (schema.ts or schema.json), share the URL, or pipe the output into another tool.

Key Benefits

Truly bidirectional: JSON Schema → Zod and Zod → JSON Schema in one tool
Built for LLM structured outputs and function calling, with a one-click strict mode
Maps formats, enums, const, unions, intersections, nullable, defaults, descriptions, and $ref
Reverse direction parses Zod with a real tokenizer — never eval, never executes your code
Never throws: invalid input degrades to a safe result with an inline // note explaining why
Useful, copy-ready examples: a User object, an API response, an LLM tool call, and a $defs schema
Runs entirely client-side — schemas never leave your browser; shareable URL + pipe-to-tool output

Common Use Cases

Turning an OpenAI/Anthropic function-call JSON Schema into a Zod validator for the model’s response
Generating a strict JSON Schema from an existing Zod schema to use as a tool/function definition
Migrating backend JSON Schema contracts into front-end Zod validation without hand-translating
Keeping a shared data shape in sync across a TypeScript app and a JSON-Schema-driven API
Quickly sanity-checking how a JSON Schema construct (const, anyOf, $ref) maps onto Zod

JSON Schema

Paste a JSON Schema (draft 2020-12 / OpenAPI-style). $ref into $defs / definitions is resolved inline.

Options

Schema / const name

Used as export const <name> = …

Export inferred type (z.infer)

Also emits: export type Name = z.infer<typeof Name>

One shape, two homes

LLM apps keep the same data shape in two places: a JSON Schema for the model’s function-calling / structured-output request, and a Zod schema to validate the reply in TypeScript. This converter goes both ways — formats, enums, const, unions, $ref, nullable, defaults and descriptions included — so you never hand-translate. The reverse parser supports a documented Zod subset: z.object z.string z.number z.boolean z.null z.array z.enum z.literal z.union .optional() .nullable() .min(n) .max(n) .int() .email() .url() .uuid() .datetime() .default(x) .describe(...) .regex(/.../) .strict().

JSON SchemaZod

Zod schema

Drop into your TypeScript app to validate model output

import { z } from 'zod';

export const User = z.object({
  id: z.string().uuid(),
  name: z.string().min(1).max(80),
  email: z.string().email(),
  age: z.number().int().min(0).max(120).optional(),
  role: z.enum(['admin', 'editor', 'viewer']).default('viewer').optional(),
  isActive: z.boolean().describe('Whether the account is enabled').optional(),
  tags: z.array(z.string()).max(10).optional(),
}).strict();

export type User = z.infer<typeof User>;

Was this tool helpful?

Share Your Experience

Help others discover this tool!

Share on

Related guide

Guide · 14 min readPublic Form APIs: What You Can and Can't Trust From the BrowserYour form is a suggestion; your endpoint is the contract. A practical guide to securing contact forms, newsletters, comments, and feedback endpoints: server-side validation, email header injection, honeypots and rate limiting, anchored origin checks, client-supplied identity, and double opt-in.Read the guide

JSON Schema ⇄ Zod Converter

How to Use

Key Benefits

Common Use Cases

JSON Schema

Options

One shape, two homes

Was this tool helpful?

Share Your Experience

Related guide

Related tools

Frequently Asked Questions - JSON Schema ⇄ Zod Converter