Server has 23 tools with basic schemas and descriptions, but significant quality gaps. All tools have descriptions (10-50 chars, mostly terse), and input schemas are present with type definitions. However, descriptions are uniformly brief and lack actionable context for LLM selection. No output schemas documented. Parameter descriptions are minimal. Error handling is absent from visible code. Tool naming is inconsistent (verb_noun for some, but 'eval', 'snapshot', 'wait' lack clarity). No tool annotations (readOnlyHint, destructiveHint). Session parameter is overloaded across all tools, forcing agents to track session state manually. No pagination support despite tools returning lists (tab_list, network_requests, session_list). Critical tools like 'run_raw_command' expose arbitrary command execution with minimal validation.

• Document output schemas for all tools. E.g., tab_list should return {tabs: [{id: string, url: string, title: string}], total: number}. Include field descriptions so LLMs know what to extract.
• Expand descriptions to 50-150 chars with actionable context. E.g., 'Get visible text from the page (or a specific element if selector provided). Use before clicking to verify page state. Returns plain text, not HTML.'
• Add tool annotations: mark get_* and read-only tools with readOnlyHint=true; mark session_close, eval, click, fill, type with destructiveHint=true; mark idempotent tools with idempotentHint=true.
• Refactor session management: provide session_create and session_info tools to make session lifecycle explicit. Document session defaults and cleanup behavior.
• Add input validation and error recovery guidance. E.g., for run_raw_command, validate args array length and return 'Invalid args: must be non-empty array of strings. Example: ["command", "--flag", "value"]' on failure.