This server exhibits severe definition quality gaps across all dimensions. Tool descriptions are present but generic and lack LLM-optimized guidance. Input schemas are visible but incomplete: most parameters lack type definitions (e.g., 'dropdownOptions' and 'choiceOptions' in process_question have no item types). No output schemas are documented. Parameter descriptions are minimal (10-30 chars) and fail to explain constraints, formats, or dependencies. Error handling is absent, no recovery guidance, no categorization, no actionable messages. The server appears designed for a specific exam-cheating use case rather than general-purpose agent integration, with tools like 'process_question' and 'cdp_mouse_click' that combine multiple concerns and lack idempotency guarantees.

• Document output schemas for all tools. E.g., gemini should return {text: string, tokens_used: number, model: string}. capture_tab should return {image_base64: string, width: number, height: number}.
• Add type definitions to array parameters. E.g., 'dropdownOptions': {type: 'array', items: {type: 'string'}, description: 'List of available dropdown options (1-100 items, max 255 chars each)'}.
• Expand parameter descriptions to 50-150 chars with constraints. E.g., 'x': {type: 'number', description: 'X coordinate in pixels (0-9999, relative to screen origin at top-left)'}.
• Implement error handling with recovery guidance. Return structured errors: {error: string, code: string, retryable: boolean, suggestion: string}. E.g., 'API key invalid. Verify GEMINI_API_KEY environment variable is set.'
• Move all credentials to server-side injection. Never expose GEMINI_API_KEY, AZURE_SPEECH_KEY in tool parameters or responses. Use environment variables or vault only.