What this does

JSON Schema defines the structure and types of parameters for tool-calling functions. Well-defined schemas help the LLM generate correct arguments and reduce parsing errors.

Steps

Write a minimal schema. Every tool schema needs name, description, and parameters.

tool_schema = {
    "type": "function",
    "function": {
        "name": "send_email",
        "description": "Send an email to a recipient",
        "parameters": {
            "type": "object",
            "properties": {
                "to": {
                    "type": "string",
                    "description": "Recipient email address"
                },
                "subject": {
                    "type": "string",
                    "description": "Email subject line"
                },
                "body": {
                    "type": "string",
                    "description": "Email body content"
                }
            },
            "required": ["to", "subject", "body"]
        }
    }
}

Include type-rich properties. Use enum, integer, array, and number types to constrain values.

{
    "name": "search_database",
    "parameters": {
        "type": "object",
        "properties": {
            "query": {"type": "string"},
            "limit": {"type": "integer", "minimum": 1, "maximum": 100},
            "sort_order": {"type": "string", "enum": ["asc", "desc"]},
            "filters": {
                "type": "array",
                "items": {"type": "string"}
            }
        },
        "required": ["query"]
    }
}

Add descriptions for every field. The LLM uses descriptions to understand parameter semantics.

properties = {
    "temperature": {
        "type": "number",
        "description": "Temperature in Celsius, e.g., 22.5"
    },
    "unit": {
        "type": "string",
        "enum": ["celsius", "fahrenheit"],
        "description": "Temperature unit"
    }
}

Generate schema from Python functions automatically. Use LangChain's @tool decorator.

from langchain.tools import tool

@tool
def fetch_user(user_id: int, include_deleted: bool = False) -> dict:
    """Fetch a user by ID. Optionally include deleted users."""
    return {"id": user_id, "deleted": include_deleted}

print(fetch_user.args_schema.schema())
# Auto-generated JSON Schema

Validate arguments against schema. Use jsonschema library to validate before execution.

from jsonschema import validate, ValidationError

try:
    validate(instance=args, schema=tool_schema["function"]["parameters"])
except ValidationError as e:
    print(f"Invalid arguments: {e.message}")

Verification

python -c "
from langchain.tools import tool
@tool
def add(a: int, b: int) -> int:
    '''Add two numbers.'''
    return a + b
s = add.args_schema.schema()
print(s['properties']['a']['type'])
# Expected: integer
"

Common failures

Missing required field. Without required, the LLM may omit essential parameters, causing runtime errors.
type: object without properties. An empty properties object is valid but useless. Always define at least one property.
Enum values don't match case. The LLM may send "ASC" while the enum expects "asc". Use lowercase or handle normalization in the function.
Version mismatch - The installed package or runtime differs from the command shown; check the version first and rerun the smallest verification command.
Local environment drift - Another service, virtual environment, model, or path is being used; print the active binary path and configuration before changing the guide steps.

Related guides

How to Use OpenAI Function Calling with Tools
How to Enable Function Calling in Local Models (Ollama)