Server provides 5 tools with basic descriptions and input schemas. Naming follows verb_noun convention (search_, get_, list_, refresh_). However, descriptions are generic and lack actionable context for LLM selection. Parameters have types but descriptions are minimal (e.g., 'Search keyword' without format guidance). Output schemas are not documented, tools return JSON strings without field-level documentation. Error handling is present but generic ('Model not found'). No parameter validation constraints (enums, ranges). Missing idempotency guidance for refresh_fal_models. Overall: functional but below production baseline.

• Document output schemas for all tools. Specify fields returned (model_id, title, category, api_docs, pricing) with types and descriptions so LLMs know what to extract.
• Add enum constraint for category parameter in list_fal_category. Replace hardcoded CATEGORIES list with formal JSON Schema enum so LLMs pick valid options without hallucination.
• Enhance parameter descriptions with format and range guidance. E.g., 'Search keyword (1-50 chars, alphanumeric + hyphens). Use search_fal_models for keyword search or list_fal_category to browse by category.'
• Add recovery guidance to error messages. E.g., 'Model "xyz" not found. Try search_fal_models("xyz") to find similar models, or call get_fal_categories() to see available categories.'
• Document refresh_fal_models as a WRITE operation with side effects. Add note: 'This tool re-crawls fal.ai (5-10 sec, rate-limited). Use sparingly. Returns error list if any category fetch fails. Idempotent: safe to retry.'