Both tools have empty input schemas (no properties defined) and minimal descriptions lacking parameter guidance. run_code accepts no documented parameters despite being a code execution tool, critical for safety. get_selection_as_image similarly has no schema detail. Descriptions are present but generic (34-92 chars), below the 50-200 char LLM-optimized baseline. No output schemas documented. No error handling guidance. This is a proxy server that delegates to Sketch's native MCP, but the placeholder tool definitions are severely underdeveloped.

• Add explicit input schema to run_code: document 'code' parameter (string, required), execution context (Sketch API version, available globals), timeout, and output format. Example: {"type": "object", "properties": {"code": {"type": "string", "description": "ECMAScript 2020 code to execute in Sketch context. Has access to sketch global and plugin API."}}, "required": ["code"]}
• Document output schemas for both tools. run_code should return {"success": boolean, "result": any, "error": string}. get_selection_as_image should return {"imageData": string (base64), "format": "png", "width": number, "height": number}.
• Expand run_code description to 100-150 chars: 'Execute ECMAScript 2020 code in Sketch plugin context. Returns result or error. Use for automating Sketch operations. Code has access to sketch global and plugin API.'
• Add error handling guidance: 'If code execution fails, return error message with line number and suggestion. If Sketch is not running, return clear 'Sketch unavailable' error, not generic placeholder.'
• Distinguish placeholder responses from real tool responses. When Sketch is unavailable, return error with status 'unavailable' rather than returning tool definitions that look identical to working ones.