VPS Browser MCP has critical definition gaps. Tool descriptions are present but minimal (10-30 chars), falling well below the 50-200 char baseline for LLM-optimized descriptions. Input schemas exist for most tools but lack depth: parameters have type and description, but no enums, ranges, or validation constraints. Output schemas are completely undocumented, LLMs cannot infer what fields to expect from responses. Error handling is absent; no recovery guidance or error classification. The execute_js tool is particularly dangerous with no safety constraints or warnings. Tool composition is reasonable (6 focused tools), but naming could be clearer (e.g., 'get_dom' vs 'get_page_html'). No tool annotations (readOnlyHint, destructiveHint) despite clear risk levels.

• Expand all tool descriptions to 50-200 characters. Example: 'get_url' → 'Retrieve the current tab's URL and page title. Use this to verify navigation or extract the current page context before taking actions.'
• Document output schemas for all tools. Specify what fields each tool returns (e.g., get_url returns {url: string, title: string}), data types, and pagination if applicable.
• Add parameter constraints: enums for selector (CSS selector format), text (max length), js (code validation hints). Replace free-form strings with validated inputs.
• Implement error handling with recovery guidance. Example: 'Selector not found. Verify the CSS selector is valid and the element exists on the page. Call get_dom() first to inspect available elements.'
• Add tool annotations to the MCP schema: mark click/type/execute_js with destructiveHint=true, get_url/get_dom/screenshot with readOnlyHint=true. This enables safer agent planning.