Two tools with reasonable structure but meaningful gaps. web_search has a clear action verb and multi-parameter input schema; use_gemini is straightforward. Descriptions are present and moderately detailed (177 and 217 chars respectively), but lack explicit guidance on WHEN to use each tool vs alternatives, dependencies, or recovery paths. Input schemas are well-formed with types and reasonable descriptions. However, output schemas are not documented, tools return Union types but no explicit structured output spec is visible. Error handling is absent, no guidance on retryability, fallback behavior, or what to do if Gemini API fails or web search returns no results. Parameter descriptions could specify constraints more formally (e.g., query length limits, model availability windows). The server is HTTP/Streamable HTTP (good) but lacks tool annotations (readOnlyHint, idempotentHint) that would clarify intent.

• Document output schemas explicitly. Create a schema definition (even as a comment) showing what fields web_search returns (text, web_search_queries, citations) and use_gemini returns (text). Include field types and a sample response.
• Add error handling guidance to descriptions. For web_search: 'If no results are found, try rephrasing the query with different keywords or breaking it into smaller sub-queries.' For use_gemini: 'If the response is truncated or the model returns an error, try shortening the prompt or splitting the request into smaller steps.'
• Declare readOnlyHint=true and idempotentHint=true on both tools (via tool annotation comments or fastmcp plugin configuration) to signal safety to agents.
• Enhance descriptions to clarify when to use each tool: 'Use web_search for current events, recent news, real-time data, or any time-sensitive information. Use use_gemini for reasoning, planning, content generation, or when you need model knowledge without live search.' Add a note: 'If you need both live search AND reasoning, consider calling web_search first, then use_gemini on the results.'
• Add constraints to parameter descriptions. E.g., query: 'The search query (max 500 characters, use natural language).' prompt: 'The task or question (max 4000 characters; keep concise for faster responses).'
• Document result limits. For web_search with include_citations=true: 'Response may include up to 10 citations; if more are needed, refine the query.' For use_gemini: 'Response is typically 100 - 2000 chars; if truncated, ask a follow-up or split the request.'