This server has three tools with significant definition quality gaps. Tool names lack strong action verbs and are somewhat unclear (e.g., 'run_lsd' is domain-specific jargon). Descriptions are present but generic and lack LLM-optimized guidance on WHEN to use each tool. Input schemas use Zod but lack descriptions for parameters, the 'query' and 'code' parameters have minimal context. Output schemas are not documented anywhere in the source. The 'run_lsd' tool is particularly problematic: it accepts arbitrary code execution with WRITE risk but provides no validation guidance, error recovery instructions, or safety warnings. No tool provides pagination, structured output documentation, or error handling guidance.

• Rename 'run_lsd' to 'execute_lsd_code' and add a critical safety warning: 'Executes arbitrary LSD code with potential side effects. Only use for trusted, reviewed code. Returns a structured array of objects with keys dependent on the query.'
• Document the output schema for each tool as JSON Schema. E.g., brave_search returns [{result_link: string, title?: string, snippet?: string}]. run_lsd returns Array<Record<string, any>> where keys vary by query.
• Rename 'find_trips' to 'search_available_tools' or 'discover_lsd_packages' to reduce domain jargon. Split registration logic into a separate 'register_tool' function if possible.
• Add parameter descriptions with constraints: For 'query' params, clarify expected format (natural language keywords? boolean operators?). For 'code' param, explain valid LSD syntax or reference docs.
• Add per-tool guidance on when to use them: 'Use brave_search to find external web resources when tools fail. Use run_lsd to transform or process data with custom logic. Use search_available_tools to discover domain-specific tools before proceeding.'
• Add error handling guidance: 'If brave_search fails, try a simpler query or use run_lsd to fetch HTML directly. If run_lsd fails with parse error, check syntax against LSD docs resource. If search_available_tools returns no results, use run_lsd with a discovery query.'
• For brave_search, return a structured array with consistent field names (url, title, snippet) instead of JSON-stringified text. This enables LLM chaining to downstream tools.