Server has 6 tools with mixed quality. Most filter tools (apply_gaussian_blur, apply_sharpen, apply_high_pass, apply_smart_blur) have minimal descriptions (10-80 chars, below the 194-char baseline) and sparse parameter documentation. apply_filter lacks parameter constraints and guidance on valid method names. start_server is a lifecycle tool with adequate schema but unclear purpose in an MCP context. Schemas are present but descriptions are consistently underdeveloped for LLM reasoning. No error handling guidance, no output schemas documented, and no tool annotations (destructiveHint/idempotentHint) despite all tools being WRITE operations.

• Expand all tool descriptions to 100-200 chars. Include WHAT the tool does, WHEN to use it (e.g., 'Use apply_gaussian_blur to soften focus or reduce noise'), and WHAT it returns (e.g., 'Returns success status and modified layer metadata'). Example: 'Apply Gaussian blur to the active layer to soften focus or reduce noise. Commonly used for background blur or noise reduction. Returns success status and the modified layer ID.'
• For apply_filter, replace free-form 'method' parameter with an enum of supported filter names (e.g., 'apply_gaussian_blur', 'apply_sharpen', 'apply_high_pass') or add a discovery tool (list_available_filters) that returns valid method names. Document the expected return value.
• Add tool annotations to all WRITE tools: set destructiveHint=true and idempotentHint=false (or true if filters are idempotent). This signals to agents that these operations modify state and may need confirmation.
• Document output schemas for all tools. Example for apply_gaussian_blur: 'Returns {success: boolean, layer_id: string, message: string}'. Include error cases: 'On failure, returns {success: false, error: string, recovery_hint: string}'.
• Add parameter constraints: for radius/threshold parameters, specify min/max values (e.g., 'radius: 0.1 - 100 pixels'). For apply_sharpen, document whether it accepts parameters or operates with fixed settings.