Single tool 'generateQuestions' has a basic description but critically lacks a complete input schema in the visible source. The description is adequate (89 chars) but the schema shows only 'count' parameter with minimal type info. No output schema is documented. The tool name lacks a clear action verb prefix (should be 'generate_questions' per verb_noun convention). No error handling guidance, no parameter constraints (e.g., min/max for count), and no documentation of the response structure. The implementation is inferred from llmService.ts rather than explicitly registered in an MCP server definition file, which caps the score further.

• Create an explicit MCP server file (e.g., src/mcp-server.ts) that registers tools via @modelcontextprotocol/sdk with full schema definitions, not inferred from service files.
• Add numeric constraints to 'count' parameter: min=1, max=100, default=10. Document in description: 'Number of questions to generate (1 - 100, default: 10).'
• Document the output schema explicitly: 'Returns an array of Question objects, each with: id (number), text (string), category (string, optional). Example: [{"id": 1, "text": "...", "category": "Personal"}]'
• Rename tool to 'generate_questions' (snake_case) and update description to include error recovery: 'If generation fails, the agent will fall back to local question templates.'
• Add parameter validation and error messages: if count < 1 or > 100, return 'Invalid count: must be 1 - 100' instead of silent failure.