MCP (Model Context Protocol) サーバーのディレクトリ&稼働監視 API。GitHub から公開 MCP サーバーを自動収集し、毎時ヘルスチェックした結果を提供します。
MCPHub API has a mixed quality profile. The 10 tools show reasonable naming (verb-prefixed) and descriptions in Japanese, but several critical gaps emerge: (1) Schema completeness varies, some tools lack parameter type definitions; (2) descriptions are brief but present; (3) error handling is basic with no recovery guidance; (4) no input validation hints in parameter descriptions; (5) parameter descriptions lack format/range constraints (e.g., 'email' param in register has no format validation guidance); (6) output schemas are not explicitly documented in the source; (7) no tool annotations (readOnlyHint, destructiveHint, idempotentHint). The server appears well-structured with FastAPI and proper HTTP transport, but falls short of production-grade tool patterns. Strengths: all tools have names and descriptions; some parameters show constraints (ge=1, le=100); authentication via API key is sound. Weaknesses: no explicit output schemas; no error guidance in descriptions; missing parameter-level format hints.
ヘルスチェック履歴取得
MCP サーバー詳細取得
API利用状況確認
サービスヘルスチェック
MCP サーバー・Claude Skills 一覧取得
APIキー発行(無料)
API 情報
Output schemas are not documented for any tool. LLMs cannot plan downstream calls or extract correct fields without knowing response structure.
Parameter descriptions lack explicit format/range/enum constraints. Validation rules (VALID_CATEGORIES, VALID_SORT, VALID_HEALTH, VALID_TOOL_TYPES) exist in code but are not reflected in parameter descriptions, forcing LLMs to guess valid values.
Tool descriptions lack recovery guidance or error context. Error responses in code (e.g., HTTPException 400, 404, 503) provide no hint to LLMs about what to do next, search users, retry, or ask user for clarification.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | B | 71 | 2026-07-28+ | v2 |
| 2026-03-09 | C | 60 | - | v1 |
GitHub APIクローラー起動(管理者専用)
ヘルスチェック起動(管理者専用)
スコア再計算(管理者専用)
Generic or vague tool names reduce clarity. 'root' should be 'get_api_info'; 'trigger_score_update' should be 'recalculate_quality_scores'; 'trigger_crawl' should be 'start_github_server_crawl'. LLMs conflate similar tools and misuse them.
'register' tool description does not explain what happens (API key issued), when to call (once, for signup), or what comes next (use X-API-Key header). Missing critical context for LLM selection.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Admin tools (trigger_crawl, trigger_health_check, trigger_score_update) should be marked destructiveHint=true or indicate side effects explicitly.
Parameter 'email' in register tool has no format hint (should state 'valid email, format: RFC 5322'). Input validation rules in code exist but are not surfaced in descriptions.
get_health_history and other discovery tools lack dependency hints. Description should explain: 'Call get_server() first to verify server exists, then call this to fetch health timeline.'