This server has exactly one tool (get_pitch) with minimal definition quality. The tool has a valid name starting with a verb (get_), a brief description, and a single input parameter with type and description. However, the definition lacks output schema documentation, has no error handling guidance, no parameter constraints, and the description is vague about the tool's purpose and when to use it. The implementation is a hardcoded stub that returns a template string, not a real service integration. The tool description 'Generate a defensive pitch to counter a competitor's advertising hook or objection' is 89 characters, below the p10 baseline of 34 but sparse on actionable detail (missing 'WHEN to use' and 'WHAT it returns'). No parameter validation, no structured output documentation, no recovery guidance, and no security considerations. This is far below production quality.

• Document the output schema with explicit field names and types. Example: {"defense_pitch": string, "confidence_level": number (0-1), "source_strategy": string}. Cite pattern:tool and rubric D (SCHEMAS & OUTPUT).
• Expand tool description to 100-200 characters. Include: (1) what the tool does, (2) when to call it (e.g., 'Use when a competitor's marketing hook threatens your pitch'), (3) what it returns (a counter-argument or defensive strategy). Reference pattern:tool-description baseline of 194 chars (p10=34, p90=392).
• Add constraints to the 'competitor_hook' parameter. Example: {"type": "string", "description": "The competitor's advertising hook or objection. Free-form text, max 500 chars. E.g. 'Our product is 50% cheaper' or 'We have 24/7 support'.", "maxLength": 500}. Reference pattern:constrained-input.
• Implement real error handling. If the competitor hook is empty or malformed, return: {"error": "Invalid competitor_hook: must be non-empty string (max 500 chars). Example: 'We provide unlimited support'.", "recovery": "Please rephrase the competitor's hook and retry."}. Reference pattern:recovery-guide and rubric E (ERROR HANDLING).
• Replace hardcoded stub with real logic (e.g., call an LLM API, consult a knowledge base, or pattern-match against known competitor tactics). Example: invoke OpenAI GPT to generate a contextual counter-pitch given the competitor's hook and product domain.
• Add tool annotations via toolAnnotations (if using MCP 2026-07-28). Mark as {"readOnlyHint": true} since generating a pitch is a read-only operation with no side effects. Reference rubric protocol section on current patterns (tool annotations).