JSON to LLM Tool Schema: Build Function Calling Definitions

Converting JSON into an LLM tool schema is the fastest way to define a function for AI APIs without writing the parameters block by hand. You paste an example of the data your function expects, and you get back the JSON Schema definition that models like Claude and GPT use to call your tool. Here is what that schema is and how to generate one.

What an LLM tool schema is

When you give a model a tool to call, you have to describe the inputs it accepts. That description is a JSON Schema object, usually with three parts:

• type: almost always object, since tool inputs are key/value arguments.
• properties: each argument, its type, and a short description the model reads to decide what to pass.
• required: the list of properties the model must supply.

The model never sees your code. It only sees this schema, so a clean, accurate schema is what makes function calling reliable. Writing it from scratch is tedious and easy to get wrong, especially with nested objects and arrays.

From JSON example to schema

The shortcut is to start from a concrete example. If you already know what a valid argument object looks like, the structure of the schema is implied: keys become properties, values reveal their types, and nesting carries over. The JSON to LLM Tool Schema tool reads your example and produces the matching parameters definition, so you spend your time refining descriptions instead of typing braces.

This is handy whether you are defining tools for the Anthropic API, OpenAI function calling, or any framework that expects standard JSON Schema parameters.

Generate your tool schema in three steps

1. Open the JSON to LLM Tool Schema tool and paste a JSON example of your function’s arguments.
2. Let it infer the schema: an object type with each key mapped to a property and its type.
3. Add descriptions to each property, set which fields are required, and copy the result into your tool definition.

Everything runs client-side in your browser. Your JSON never leaves your device and nothing is sent to a server, so it is safe to use with example payloads that mirror real data.

Tips for a schema the model will use well

• Write a clear description for every property. The model leans on these to fill arguments correctly.
• Keep required honest. Only list fields the function truly cannot run without.
• Use enums for fixed choices so the model picks valid values instead of guessing.

Related tools

• LLM Token Counter & Cost Estimator: tool schemas count toward your prompt, so check how many tokens they add.
• llms.txt Generator: describe your whole site for AI agents, not just one function.
• Block AI Crawlers (robots.txt): control which AI bots are allowed to read your site.

Paste your JSON, refine the descriptions, and drop a working tool schema straight into your next AI call.