mpget is a STDIO-only MCP server with 3 tools exposed via the MCP protocol. Tool definitions have schemas and descriptions, but exhibit several quality gaps: (1) naming lacks consistent verb-noun pattern (search, album, content are weak verbs or nouns); (2) descriptions are brief but present (65-85 chars); (3) parameter descriptions are present but some are Chinese-only; (4) output schemas are not explicitly documented in the MCP tool definitions; (5) error handling in the implementation provides structured JSON responses but LLM recovery guidance is missing. The tools are read-only with no destructive operations, which reduces risk. Overall, the server is functional but mediocre, comparable to community-average tools scoring 50-60.

• Rename tools to verb_noun pattern: 'search' → 'search_articles', 'album' → 'get_album_articles', 'content' → 'fetch_article_content'. This improves LLM intent inference.
• Translate all parameter descriptions to English. Add brief constraint hints: 'page: integer, ≥1, default 1', 'maxPages: 1 - 100, default 5'.
• Document output schemas in tool definitions. For 'search': 'Returns { results: Array<{title: string, url: string, preview?: string, real_url?: string}>, page: number }'. For 'album': 'Returns { articles: Array<{title, url, ...}>, next_msgid?: string, next_itemidx?: string }'. For 'content': 'Returns { url: string, content: string, error?: string }'.
• Expand tool descriptions to 100 - 200 chars each. Example for 'search_articles': 'Search WeChat Official Accounts (公众号) for articles matching a keyword. Returns article metadata with optional real URLs (bypasses WeChat redirect). Use resolve=true to fetch actual article links instead of WeChat Share URLs.'
• Add recovery guidance to error messages. E.g., if album fails with ALBUM_RET_5001: 'Album ID not found. Verify: 1) albumId is numeric/alphanumeric 2) biz value matches article URL (var biz in page source) 3) Account has published this collection.'
• Clarify parameter dependencies. Note that 'all=true' enables 'maxPages', and 'resolve=true' adds ~5 - 10s latency per result. Add to descriptions: 'Note: resolve=true requires fetching each article URL separately (~5 - 10s per 10 results).'
• Consider adding a 'limit' or 'max_results' parameter to 'search' to cap response size. Currently unbounded pagination could exhaust context.