Six tools with minimal, demo-focused descriptions that lack LLM-optimization guidance. Tool names are simple verbs (add, subtract, greet, ping, hello, code_review) but descriptions are explicitly marked 'Demo' and discourage real use, signaling this is a reference implementation, not production-grade. No input schemas visible in source code, only parameter names and types inferred from the tool list. No output schemas documented. Error handling and recovery guidance absent. Naming is clear but descriptions fail the 50 - 200 char LLM-optimization baseline (most are 100 - 180 chars but include anti-patterns like 'do not use for real math').

• Remove 'Demo' and 'do not use' language from descriptions. Rewrite as: 'add: Adds two integers and returns the sum. Use to verify arithmetic operations in integration tests.' This signals intent without discouraging use.
• Document output schemas for each tool. E.g., add returns {result: integer, operation: string}. code_review returns {feedback: string, issues: array}. This enables LLM planning.
• Add explicit JSON Schema definitions with type constraints. For greet, add an enum for greeting: ['Hello', 'Hi', 'Welcome', 'Greetings'] or accept any string. For add/subtract, specify integer ranges (e.g., -1000000 to 1000000).
• Expand parameter descriptions with format and constraint hints. E.g., 'name: Person to greet (string, 1 - 100 chars, alphanumeric + spaces)' instead of just 'Person to greet (required)'.
• Add error handling examples. E.g., 'If name is empty, returns error: "name required (1 - 100 chars)". If a or b exceed ±1M, returns error: "integer out of range".' This guides LLM recovery.