Three tools with basic schemas and descriptions, but significant gaps in parameter documentation, output schema clarity, and error handling guidance. Tool names follow verb_noun convention (generate_*, check_*), which is good. However, descriptions lack actionable context for LLM selection, parameter descriptions are minimal, and output schemas are not formally documented. No input validation guidance, no error recovery hints, and no structured output specification. Baseline: avg tool description ~100 chars (below 194 baseline); most params lack detailed constraints or format guidance.

• Document output schemas explicitly. Return structured JSON objects: {task_id: string, status: string, message: string} for generate_* tools; {task_id: string, status: 'pending'|'completed'|'failed', result_urls: string[], error_message?: string} for check_task. This enables LLMs to reliably extract and chain results.
• Add enum constraints to parameters. Convert 'duration' description to schema with enum: [5, 10]; convert 'aspect_ratio' to enum: ['16:9', '9:16', '1:1']; convert 'task_type' to enum: ['video', 'image']. Remove free-form strings that invite hallucination.
• Expand tool descriptions to 100-200 chars with WHEN/WHY context. Example: 'Generate a video from text or image using Kling AI. Use for text-to-video (mode=text) or image-to-video (mode=image). Returns task_id; poll with check_task() to monitor progress. Requires KLING_API_KEY environment variable.'
• Implement error classification and recovery hints. Distinguish: 'API error: 401 Unauthorized, check KLING_API_KEY environment variable' (user-fixable) vs 'API error: 429 rate limit, retry in 60 seconds' (retryable) vs 'API error: 500 internal server error, contact Kling support' (fatal). Return structured error: {error: string, category: 'auth'|'rate_limit'|'validation'|'server', retry_after?: number}.
• Add parameter validation and bounds. Validate prompt length (e.g., max 1000 chars), quantity range (1-4), duration enum (5|10). Return clear errors: 'Invalid quantity: 5, must be 1-4' instead of silent API failures.