Function Call Mock Response Generator

Function call mock generator

Building the code that handles an LLM’s tool calls means you need example tool-call payloads to test against — but calling the real API for each one is slow and costs tokens. This tool reads your function/tool JSON Schema and synthesises type-correct mock arguments, giving you instant fixtures for your parsing and handler logic.

How it works

The input is parsed as JSON and the tool locates the parameters schema — under function.parameters (OpenAI), input_schema (Anthropic), or at the top level (bare JSON Schema). It then walks the schema recursively:

• enum fields take their first option (real, allowed values)
• Strings derive a plausible value from the property name — an email field becomes [email protected], a userId becomes user_123
• Numbers take the midpoint of minimum/maximum if declared, otherwise default to 0
• Booleans default to true
• Arrays produce two sample items of the declared item type
• Nested objects recurse through all their properties

Generation is deterministic: the same schema always yields the same mock. The result is wrapped in a realistic tool-call envelope — including id, type, and function.name — that you can paste straight into a test fixture.

Worked example

For an OpenAI-style schema like this:

{
  "name": "create_order",
  "parameters": {
    "type": "object",
    "properties": {
      "product_id": { "type": "string" },
      "quantity":   { "type": "integer", "minimum": 1, "maximum": 100 },
      "priority":   { "type": "string", "enum": ["low", "normal", "high"] }
    },
    "required": ["product_id", "quantity"]
  }
}

The generator produces a mock tool call like:

{
  "id": "call_mock_abc123",
  "type": "function",
  "function": {
    "name": "create_order",
    "arguments": "{\"product_id\":\"product_id_value\",\"quantity\":50,\"priority\":\"low\"}"
  }
}

You can feed this directly into the handler under test, or adapt the values by hand before using it as a fixture.

When this is most useful

• Unit-testing handler functions that parse arguments strings and call downstream services — no API tokens needed.
• Integration testing schemas across OpenAI and Anthropic formats without maintaining two separate fixture files.
• Rapid prototyping when you want to see what a handler receives before the real model is wired in.
• CI/CD pipelines where deterministic, cost-free fixtures prevent flaky tests and token spend from test suites.

Tips and limitations

• Every property is included. The generator emits all defined properties, not just the required ones, so the mock exercises your full parsing path.
• Use enums for realistic values. The generator picks the first enum option — a real, allowed value — making the mock more meaningful than a generic placeholder.
• It validates structure, not semantics. The mock satisfies the schema’s type constraints; it will not invent business-meaningful data. Adjust specific values by hand where your tests assert on exact content (for example, a specific order ID format).