Skip to content

What is an agent tool schema and how do parameters get validated before invocation?

Beginner · 2 min read · AI Agents & MCP

Open in study mode to save this question and track progress

Short answer: An agent tool schema is a formal JSON Schema contract defining a tool's function name, natural language description, and required parameter types; pre-invocation validation uses validation engines (like Pydantic) to type-check LLM-generated arguments before executing underlying Python code.

Detail

LLMs generate tool calls as raw JSON strings. Without strict schema validation, invalid parameter types (e.g. string passed instead of int) cause runtime crashes.

LLM Function Call ──► JSON Arguments ──► [Pydantic Validator] ──► Execute Tool Python Function
                                                 │
                                                 ▼ (If Validation Fails)
                                        Return Error to Agent

Core Elements of Tool Schemas

  1. Description String: Natural language explanation instructing the LLM when and why to choose the tool.
  2. Property Schema: JSON Schema data types (string, integer, boolean, array, enum).
  3. Required List: Array of mandatory parameter names.

Example

Pydantic tool schema definition and validation in Python:

python
from pydantic import BaseModel, Field, ValidationError

class SQLQueryToolSchema(BaseModel):
    query: str = Field(description="ANSI SQL SELECT query to execute")
    limit: int = Field(default=100, ge=1, le=1000, description="Max rows to return")

def validate_and_execute(raw_args: dict):
    try:
        validated_args = SQLQueryToolSchema(**raw_args)
        return f"Executing valid SQL query (limit {validated_args.limit}): {validated_args.query}"
    except ValidationError as e:
        return f"Tool Argument Error: {e}"

print(validate_and_execute({"query": "SELECT * FROM users", "limit": 50}))

Interview tips

  • Emphasize that clear, descriptive tool descriptions in schemas are critical for LLM function routing accuracy.
  • Discuss handling validation errors gracefully by returning validation feedback to the LLM agent for self-correction.

Related Concepts

⬅ Back to AI Agents and MCP · All topics