The server defines 2 tools with good parameter schemas and reasonable descriptions. Tool names follow verb_noun convention (query_*, search_*). Both tools have input schemas with proper types and constraints. However, descriptions lack depth about when to use each tool vs. the other, parameter descriptions are minimal, and there is no documented output schema. Error handling returns text-wrapped errors rather than structured guidance. The resource feature is present but adds minimal value (static server info).

• Document the output schema for both tools. Example: 'Returns: { text: string (the response from Perplexity), model: string (the model used), tokens_used: integer (approximate tokens consumed) }'. This lets LLMs understand what data they receive.
• Expand model parameter description to: 'The Llama model to use. sonar (fast, web-enabled default), sonar-pro (larger, more accurate), sonar-reasoning (slower, best for complex problems). Omit to use sonar.'
• Expand temperature parameter description to: 'Response randomness (0.0-1.0, default 0.7). Use 0.0-0.3 for factual queries, 0.7-1.0 for creative tasks. Lower = more deterministic.'
• Add to search_perplexity description: 'Quick search with fixed defaults (model=sonar, max_tokens=1000, temperature=0.7). Use this for simple one-off queries. For fine-grained control, use query_perplexity instead.'
• Add to query_perplexity description: 'Full-featured query allowing model selection and parameter tuning. Use when you need specific model behavior, token limits, or system context. For simple searches, query_perplexity is faster.'
• Implement structured error responses that classify errors: 'Error type=authentication_failed message=Invalid API key. Check PERPLEXITY_API_KEY environment variable.' Or: 'Error type=rate_limited message=Quota exceeded. Retry in 60 seconds.'
• Complete and document the prompts handler. Include 1-2 reusable prompt templates (e.g., 'research_topic', 'fact_check', 'summarize'). Each should have clear arguments and expected output shape.

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry