Critical failure: ALL 13 tools have placeholder descriptions (e.g., 'NavigateTool.shape.description.value') instead of actual text. This is a hard blocker, LLMs cannot determine when or why to select tools without real descriptions. Input schemas are present with proper types and parameter descriptions, but tool-level descriptions are completely absent. No error handling guidance, no output schema documentation, and no tool annotations. This server is non-functional for agent use.

• Replace all placeholder descriptions with real, LLM-optimized text (50 - 200 chars). Example for 'navigate': 'Navigate to a URL in the browser. Waits for page load. Returns screenshot of the loaded page.'
• Document output schema for each tool. Example: 'Returns {type: "image", data: base64_png} for screenshot, or {type: "text", content: string} for console logs.'
• Add tool annotations via MCP SDK: mark navigate/click/drag/type/selectOption as destructiveHint=true; mark screenshot/getConsoleLogs/snapshot as readOnlyHint=true.
• Clarify 'element' parameter: 'CSS selector string (e.g., "#submit-btn", ".form-input") or XPath (e.g., "//button[@id=\'submit\']")'.
• Add error recovery guidance in descriptions: 'If element not found, call snapshot() first to see current page state, then retry with correct selector.'