MCP Tool Design

A tool is only as good as its description. AI models depend on clear JSON Schema to function.

Tool Definition

• Name: Concise, snake_case (e.g., calculate_risk_score).
• Description: Detailed explanation of what the tool does and when to use it.
• Input Schema: Strict types and required fields using JSON Schema.

Execution Logic

• Safety: Validate inputs before processing.
• Feedback: Provide verbose success messages or actionable error reports.
• Idempotency: Ensure running the tool multiple times with the same input has predictable results.

Best Practices

• Schema Evolution: Maintain backward compatibility.
• Context-Awareness: Use tools to supplement the model's knowledge, not replace its reasoning.