Chrome Tools MCP has 7 well-named tools that follow verb_noun conventions (list_tabs, capture_screenshot, execute_script, load_url, capture_network_events, query_dom_elements, click_element). All tools have descriptions and input schemas with Zod validation visible in src/index.ts. However, definitions lack critical production-grade polish: output schemas are NOT documented (tools return plain text or JSON without field descriptions), parameter descriptions are present but inconsistent in depth and clarity, and error handling provides generic error messages without recovery guidance or error classification. Description lengths average ~60 chars, which is below the 194-char baseline for production tools. The code shows proper Zod schema registration (good), but MCP's structured output pattern (returning typed objects with described fields) is not applied, most tools return {content: [{type: 'text', text: ...}]} without documenting what fields the response contains. Naming is excellent (all verb-first, action-clear), but the definition completeness score is dragged down by missing output documentation and generic error handling.

• Document output schemas for every tool. Create a formal response type for each: list_tabs should return {tabs: Array<{id: string, title: string, url: string, ...}>}, capture_screenshot should return {status: string, path: string, format: string}, etc. Return structured objects instead of JSON-in-text.
• Add recovery guidance to every error. When a tab is not found, return 'Error: Tab ABC123 not found. Call list_tabs() to see available tabs.' When execute_script fails, return 'Script execution failed: <details>. Verify the script syntax and that the page is fully loaded.'
• Expand tool descriptions to 100-150 characters. For execute_script: 'Execute arbitrary JavaScript in a Chrome tab. Returns the result of the last expression. Use for querying page state (document.title), triggering actions, or extracting data. Note: runs with full page context, use carefully.' For load_url: 'Load a URL in an existing Chrome tab and wait for navigation to complete. Use to change the current page. Returns success or error message.'
• Add parameter context hints. For tabId: 'Use list_tabs() to obtain a tab ID. IDs remain valid until the tab is closed.' For script in execute_script: 'JavaScript code as a string. The last expression value is returned. Example: document.title or await fetch(...).then(r => r.json())'
• Apply tool annotations via MCP's destructiveHint and idempotentHint. Mark execute_script, load_url, and click_element with destructiveHint=true so agents know these modify state and avoid retrying blindly.