Four tools with basic descriptions and minimal parameter documentation. All tools have descriptions (10-60 chars), but descriptions lack actionable context for LLM selection. Input schemas present for AddRestaurant only; other three tools have empty input objects. No output schemas documented. No error handling guidance. No parameter descriptions for AddRestaurant's three string parameters, LLMs cannot infer what 'foodType' means vs 'location'. Tool names are verb-noun (good), but descriptions are generic and under 20 chars for three of four tools, triggering hard scoring caps. No pagination, no result limits documented, no idempotency hints. PickRandomRestaurant modifies state but description does not say so.

• Expand tool descriptions to 50 - 150 characters. Example: 'GetRestaurantsAsync' → 'Retrieve all available restaurants with name, location, and cuisine type. Use before picking a random restaurant or adding a new one.'
• Add parameter descriptions and constraints to AddRestaurant. Example: foodType should be an enum (Italian, Mexican, Thai, Chinese, Japanese, etc.) with description 'Cuisine category. Must be one of: Italian, Mexican, Thai, Chinese, Japanese, Indian, Mediterranean, American, Vietnamese, Korean.'
• Document output schemas for all tools. Example: GetRestaurantsAsync returns {restaurants: [{id: string, name: string, location: string, foodType: string, dateAdded: ISO8601}], total: number}.
• Mark PickRandomRestaurant as a write operation in its description: 'Randomly select a restaurant and increment its visit counter. This modifies state and is not idempotent.'
• Add error handling guidance. Example: 'If no restaurants are available, returns {message: "No restaurants available. Please add some restaurants first!"}. Call AddRestaurant to add options.'