greybeard exposes 4 tools via MCP with generally clear intent but inconsistent quality. All tools have descriptions and input schemas are present, but parameter descriptions vary in completeness. Tool names follow verb_noun convention (review_decision, self_check, coach_communication, list_packs). Schemas use proper JSON Schema format with types. However, several parameters lack descriptions, output schemas are not documented, and error handling guidance is absent.

• Document output schemas for all tools. For review_decision, specify: {type: object, properties: {analysis: {type: string}, risks: {type: array, items: {type: string}}, tradeoffs: {type: array}, questions: {type: array}}, ...}. For list_packs, specify: {type: object, properties: {packs: {type: array, items: {type: object, properties: {name: string, description: string, category: string}}}}}.
• Add descriptions to all parameters, especially optional ones. For 'context' in review_decision: 'Optional: additional context about this change (background, constraints, stakeholders involved)'. For 'input' in self_check: 'Optional: supporting document or draft that you want to self-check.'
• Clarify what 'staff-core' and 'mentor-mode' content packs contain. Add to description or provide a discovery tool example: 'Run list_packs to see available options and their descriptions.'
• Add error handling documentation. Example for coach_communication: 'Returns error if audience is not one of: team, peers, leadership, customer. Returns error if concern is empty.'
• For list_packs, add parameter descriptions even if empty, and specify response structure: packs (array of {name, description, category, installed_path}). Mention if results are paginated.
• Consider adding a 'mode' parameter to coach_communication with enum values and descriptions (e.g., 'collaborative', 'assertive', 'diplomatic') to give LLM more control.