The single tool 'get_browser_screenshot' has a verb-based name (positive), but critically lacks any input schema definition, parameter descriptions, and output schema documentation. The tool is registered with mcp.NewTool('get_browser_screenshot') but no schema object is provided, the Input field is empty {}. No description text is visible in the tool registration. The tool returns mcp.NewToolResultImage() with hardcoded empty string and base64 data, but the output structure is not documented for the LLM. This violates core pattern:tool and pattern:tool-description requirements.

• Add a tool description (50 - 200 chars): 'Captures a screenshot of a selected DOM element in a browser. User clicks to select element, returns base64-encoded PNG image. Requires active browser session with embed.js loaded.'
• Define input schema with explicit parameters. If the tool accepts no parameters, document this clearly: {"type": "object", "properties": {}, "required": []}. If it accepts optional parameters (e.g., element_selector, timeout_ms), add them with descriptions and type constraints.
• Document the output schema: return type is image/png as base64 string. Specify the response structure: {"type": "object", "properties": {"image": {"type": "string", "description": "Base64-encoded PNG image data"}, "element_info": {"type": "object", "description": "Metadata about selected element (tag, classes, dimensions)"}}}
• Add error handling with recovery guidance. Replace 'no browser session' with 'No active browser session. Ensure embed.js is loaded in the target page and a WebSocket connection is established. Retry after loading the page.'
• Document prerequisites and limitations in the tool description: 'Requires a browser with embed.js injected and an active WebSocket connection to localhost:8000. User must manually select an element via click. Returns null if user presses Escape or selection fails.'