This server has critical definition gaps. All three tools have minimal descriptions (10-15 chars, well below the 50-200 char baseline for LLM-optimized tools). Input schemas are present but extremely sparse, only 'sku' and 'supplier' parameters with minimal descriptions. No output schemas are documented. No error handling guidance, no parameter constraints (enums, ranges), no idempotency hints. The tool implementations are hardcoded mock returns with no real integration. Transport is unknown, preventing remote access.

• Expand each tool description to 80-150 chars. Include: what it does, when to call it, what it returns. Example: 'Check current inventory levels for a SKU. Returns available units, reserved units, and reorder point. Call before creating orders to verify stock.'
• Document output schemas explicitly. For check_inventory, specify: {sku: string, available: number, reserved: number, reorder_point: number}. For get_forecast: {sku: string, next_30d_demand: number, confidence: string}.
• Add parameter descriptions with format/constraint hints. Example: 'sku: Stock Keeping Unit (5-20 alphanumeric chars, e.g. SKU-001). Must exist in inventory system.'
• Implement error handling with recovery guidance. Return: {error: 'SKU not found: SKU-999. Try search_skus() with a partial name or check_inventory_categories() to browse available SKUs.'} instead of silent failures.
• Add idempotency hints to tool descriptions. All three are read-only, so mark them with idempotentHint: true in tool annotations (once transport is clarified).