Function Calling Schema Generator

About Function Calling Schema Generator

Fill in a function name, description and typed parameters (type / required / enum / array item type) and get copy-ready Function Calling JSON in both OpenAI tools[] and Claude input_schema dialects. Invalid names fail loudly with a one-click fix instead of silent mangling; the supported JSON-Schema subset is fully covered and its edge (nested anyOf/oneOf) is stated on the page. Generation runs entirely in your browser.

FAQ

What's the difference between the OpenAI and Claude formats?

The parameter schema core (type/properties/required) is identical; the difference is the envelope. OpenAI wraps in a tools array: { type: "function", function: { name, description, parameters } }. Claude is flat: { name, description, input_schema } — no tools wrapper, no nested function object. This tool emits both from the same form; flip the tab to compare. Both vendors weigh descriptions heavily — the more specific, the more reliable the calls.

Why force lowercase snake_case function names?

Both vendors' examples use snake_case, and some runtimes (routing layers, older SDKs) tolerate uppercase or hyphens inconsistently. To keep the generated JSON runnable everywhere, the tool validates ^[a-z][a-z0-9_]*$; type anything else and you get an explicit error plus a one-click suggestion (the suggestion filters characters — confirm it still reads well). Parameter names are looser: camelCase or snake_case, letters first.

How do enum parameters work? Are numbers supported?

Enums are open to string/number/integer only. Enter comma-separated values: strings are taken verbatim (celsius,fahrenheit); number/integer values are parsed one by one, and any non-numeric entry fails loudly instead of emitting a broken schema. The point of an enum is collapsing open-ended generation into a closed set — with one, the model all but stops inventing values for things like temperature units or sort order.