Server defines 9 tools with complete input schemas and basic descriptions. However, descriptions are generic and lack LLM-optimized guidance on WHEN to use each tool or what distinguishes them. Parameter descriptions are minimal (e.g., 'The model to use' without format/constraint details). No output schemas documented. Error handling is absent, makeLiteLLMRequest throws generic errors without recovery guidance. Tools lack idempotency hints, permission declarations, and composition guidance. Naming is clear (verb_noun pattern) but descriptions fall short of production baseline (194 chars avg; these average ~80 chars).

• Document output schemas for all tools. E.g., chat_completion should return {type: 'object', properties: {content: {type: 'string'}, model: {type: 'string'}, usage: {type: 'object', properties: {prompt_tokens, completion_tokens, total_tokens}}}}. This enables LLMs to extract and chain results.
• Expand tool descriptions to 150-250 chars with WHEN/WHY guidance. E.g., 'Generate chat completions for multi-turn conversations. Use this for dialogue, reasoning, and creative tasks. Supports streaming. Returns content, model, and token usage. Call list_models first to see available models.'
• Add format/constraint details to parameter descriptions. E.g., 'model: The model identifier (string, e.g. gpt-4, claude-3-opus). Call list_models to discover available options.' and 'temperature: Sampling temperature (0.0 - 2.0, default 1.0). Lower values (0.0 - 0.7) for deterministic tasks; higher (1.0 - 2.0) for creative generation.'
• Implement structured error responses with recovery guidance. Instead of throwing generic errors, return {error: 'model_not_found', message: 'Model gpt-5 not found', recovery: 'Call list_models to see available models', retryable: false}. This guides LLM recovery.
• Add tool annotations to the tool definitions. Set readOnlyHint: true for list_models, health_check, model_info, create_embedding, rerank. Set idempotentHint: true for read-only tools. This clarifies retry safety and side effects.