Server exposes 6 well-named, read-only tools with complete Zod schemas and reasonable descriptions. All tools follow verb_noun naming (search_servers, list_servers, get_server, discover, etc.). Descriptions are present and contextual (avg ~120 chars), meeting the 10 - 1024 char baseline. However, parameter descriptions lack actionable constraints (e.g., 'limit' has no min/max bounds; 'sort' enum values are listed but not explained when to use each). Output schemas are not documented, LLMs cannot predict response structure. Error handling is absent: no recovery guidance, no categorization of retryable vs fatal errors. All tools are READ_ONLY (safe), but the server lacks composition guidance (e.g., when to call search_servers vs discover vs list_servers). Parameter naming is clear (query, limit, category, sort, budget) but some lack type hints in descriptions (e.g., 'budget' should state 'USD, numeric').

• Document output schemas for all tools. Specify response fields, types, and nesting. E.g., search_servers returns {servers: [{id, name, trust_score, category, ...}], total_count, next_cursor}. This enables LLMs to plan downstream calls and extract data correctly.
• Add min/max bounds to numeric parameters in descriptions. E.g., 'limit: Max results (1-50, default 10)' and 'budget: Max monthly price in USD (0-10000, optional)'. Include validation in the handler to reject out-of-range values with actionable error messages.
• Implement error handling with recovery guidance. Catch API errors and return structured responses: {error: 'not_found', message: 'Server not found. Try search_servers() with a partial name.', retryable: false}. Categorize errors as retryable, user-fixable, or fatal.
• Clarify tool composition in descriptions. E.g., search_servers: 'Use for semantic search by task description.' list_servers: 'Use for browsing by category or pagination.' discover: 'Use for AI-powered recommendations with budget constraints.' This guides LLM selection.
• Add parameter type hints and units to descriptions. E.g., 'budget: Max monthly price in USD (numeric, optional)' and 'sort: Sort order (enum: relevance=best match, trust=most reliable, stars=most popular, updated=newest)'. This prevents LLM hallucination of invalid values.