JSON Schema → TypeScript Types

Generate TypeScript types from JSON Schema

The JSON Schema → TypeScript converter turns a JSON Schema object into clean, idiomatic TypeScript interface definitions. It is built for the LLM era, where function-calling tool schemas and structured-output formats are expressed in JSON Schema — generating matching interfaces lets you parse tool arguments and model responses with full type safety.

How it works

The generator walks the schema recursively. Object types become interfaces; properties not in the required array become optional (?). Arrays map to T[], string enums map to string-literal unions, and $ref pointers resolve against definitions or $defs. anyOf and oneOf render as union types. Nested anonymous objects get auto-named child interfaces so the output stays readable. Unknown or unsupported constructs degrade gracefully to unknown.

Mapping rules at a glance

A practical worked example

Take this function-calling schema for a weather tool:

{
  "type": "object",
  "properties": {
    "location": { "type": "string", "description": "City name" },
    "units": { "enum": ["celsius", "fahrenheit"] }
  },
  "required": ["location"]
}

The generator produces:

interface Root {
  location: string;
  units?: "celsius" | "fahrenheit";
}

location is non-optional (it is in required). units becomes a string-literal union from the enum values, and the trailing ? marks it optional. You can now type the argument object passed to your weather function handler:

function handleWeather(args: Root) {
  const unit = args.units ?? "celsius";
  // TypeScript knows units is "celsius" | "fahrenheit" | undefined
}

Tips and notes

If your schema uses $ref, keep the referenced definitions under $defs or definitions so they resolve. For LLM tool calling, generate the interface for your function’s parameters schema and type the arguments you receive from the model. When a field can be several shapes, oneOf produces a discriminated-union candidate you can refine. The output is plain text — copy it directly into your project.

After generating, rename Root to something descriptive and review any unknown fields — those indicate schema constructs the generator could not map, which you should type by hand.