Server has 3 tools with basic descriptions and minimal parameter documentation. All tools lack proper input schemas with type definitions and parameter descriptions. Naming violates single-responsibility principle (generate_and_save_image combines two concerns). Error handling returns generic strings instead of actionable guidance. No output schemas documented. Parameter descriptions are minimal (e.g., 'Path where to save the image' lacks format/constraint details). Credentials exposed via environment variable but not validated at tool boundary.

• Split 'generate_and_save_image' into two separate tools: 'generate_image' (returns file_id) and 'save_image' (accepts file_id + output_path). This enables agents to compose them independently and retry individual steps.
• Add explicit JSON Schema to all parameters with type, description, and constraints. Example: 'prompt' should specify min/max length; 'output_path' should specify format (absolute Windows path) and validate against path traversal.
• Replace generic error strings with structured error responses that classify errors as retryable/user-fixable/fatal and provide recovery hints. Example: 'API rate limit exceeded. Retry after 60 seconds.' instead of 'Error generating image: ...'
• Document output schemas for each tool. Example: generate_image should return {file_id: string, expires_at: ISO8601_string} instead of a bare string.
• Add input validation at tool entry points. Validate output_path is absolute, does not contain '..', and is writable before calling GigaChat API. Return clear validation errors to guide LLM self-correction.