Single tool 'analyze_images_feature' has fundamental definition quality issues. The tool name lacks a clear action verb prefix and does not clearly convey what happens when called. The description is in Chinese (non-English, problematic for multilingual LLM deployments) and is 73 characters, within acceptable range but contextually weak. The input schema IS present with proper JSON Schema structure (red_note: string, urls: array of strings), but parameter descriptions are minimal. Output schema is completely undocumented, the function returns a string but the structure and content of that string are not specified. Error handling is present in the code but not surfaced in the tool definition. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite the tool being read-only. The tool invokes an external LLM (Gemini) to analyze images, but this dependency and potential latency/failure modes are not documented in the tool description.

• Rename tool from 'analyze_images_feature' to 'analyze_images' (shorter, clearer) or 'summarize_images' (more specific about the analysis type).
• Rewrite description in English, ~80-150 characters: 'Analyze images from a Xiaohongshu post and extract location, composition, and map information. Requires GOOGLE_API_KEY environment variable.'
• Add per-parameter descriptions: red_note should specify it must be complete post text in Chinese; urls should specify max 10 URLs, accepted formats (http/https), and timeout (30s per image).
• Document output schema explicitly in the description: 'Returns analysis formatted as: **Location:** specific place name, **Image URL:** [url], **Description:** visual details, **Is Map Image:** yes/no'.
• Add readOnlyHint annotation in tool metadata (if using FastMCP's decorator syntax, use @mcp.tool(readOnly=True)).
• Document the Gemini API dependency in the description: 'Note: Requires GOOGLE_API_KEY set as environment variable. Uses Google Gemini 2.5 Pro; see rate limits at https://ai.google.dev/pricing'.
• Improve error messages to English and actionable: 'Unable to load images. Verify URLs are valid and publicly accessible (check status codes 200-206).' instead of '图片数据处理异常'.
• Add parameter constraints: urls array, min 1, max 10 items; red_note string, min 1, max 5000 characters.
• Document expected latency and costs in description: 'Analysis takes ~5-10 seconds per image. Incurs charges based on Google Generative AI pricing.'
• Implement timeout guard: tool should fail cleanly if any image takes >30 seconds to process, not hang indefinitely.