Three tools with basic descriptions and no input schemas visible. show_reaction has a category parameter with enum-like description but no formal schema constraint. list_categories and diagnose have no parameters. Descriptions are present (60-100 chars) but lack actionable detail on when to use each tool or what downstream actions they enable. No output schemas documented. Error handling in show_reaction returns structured dicts but lacks recovery guidance for LLMs. Tools are single-purpose (good) but composition is limited, no chaining IDs or references between tools.

• Add formal JSON Schema to show_reaction input: {"type": "object", "properties": {"category": {"type": "string", "enum": ["confused", "skeptical", ...], "description": "Reaction category. Call list_categories() to see available options."}}, "required": ["category"]}
• Document output schemas for all three tools. E.g., show_reaction returns {"ok": boolean, "category": string, "gif_id": string, "error": string (optional)}. This enables LLMs to extract and chain results.
• Expand show_reaction description to 100-150 chars: 'Display a sarcastic reaction GIF in a borderless popup (bottom-right, auto-closes after 4s). Use when the user's request is confusing, skeptical, or deserves a humorous response. Returns status; the popup is a side effect visible to the user.'
• Add recovery guidance to error responses. When category is invalid, return: {"ok": false, "error": "Unknown category 'xyz'. Call list_categories() to see valid options: confused, skeptical, ..."} instead of a bare error.
• Add tool annotations to the FastMCP decorator: @mcp.tool(readOnlyHint=True) for all three tools, signaling to LLMs that they are safe to call and retry without side effects.