All three tools lack input parameter schemas in the source code. While tool names follow verb_noun convention (get_*, download_*), descriptions are minimal (10-20 chars) and parameter descriptions are absent from the tool registration. The code shows SearchInput and TimeInput structs with jsonschema tags, but these are NOT wired into the mcp.Tool definitions via AddTool(). The tool registration in main.go passes only Name and Description to mcp.Tool{}, with no InputSchema field visible. This means LLMs cannot see parameter constraints, types, or descriptions at runtime. Error handling returns plain text rather than structured recovery guidance.

• Wire InputSchema into mcp.Tool definitions. Use the jsonschema tags from SearchInput/TimeInput structs to generate JSON Schema objects and pass them to AddTool(). Example: mcp.Tool{Name: 'get_search_results', InputSchema: &jsonschema.Schema{Type: 'object', Properties: {...}}}.
• Expand tool descriptions to 100-200 chars. Include WHAT (searches the web), WHEN (when you need current info), and WHAT IT RETURNS (array of {title, url}). Example: 'Search the web using Brave Search. Returns up to 10 results with title and URL. Use when you need current information not in your training data.'
• Add parameter descriptions to the schema. For 'query', add description: 'Search query string (required, 1-500 chars). Example: "latest AI news"'. For 'timezone', add description: 'IANA timezone identifier (required). Examples: America/New_York, Europe/London, Asia/Tokyo'.
• Implement actionable error messages. Instead of 'Error performing search: <err>', return 'Search failed: network timeout. Retry in 5 seconds or try a simpler query.' Categorize errors as retryable vs user-fixable.
• Document output schemas in tool descriptions and return structured JSON with typed fields. Example for get_search_results: 'Returns {results: [{title: string, url: string, snippet?: string}], total_count: number}'. Avoid returning raw HTML or unstructured text.