搜索并下载 e621 / rule34 / E-Hentai / pixiv 资源的 MCP 服务器
Media Hunter MCP demonstrates solid definition quality with comprehensive schemas, detailed parameter descriptions, and proper tool annotations. All 5 tools have explicit input schemas with type constraints and enums. Descriptions are substantive (100-300+ chars) and explain WHEN to use each tool. Output schemas are fully documented with structured POST and DOWNLOAD_DATA_SCHEMA definitions. Tool names follow verb_noun pattern (media_hunter_search, media_hunter_download). However, some parameter descriptions are verbose and could be more concise for LLM parsing. Error handling guidance is present but could be more explicit about recovery paths. Chinese descriptions may reduce LLM comprehension in some contexts.
检查各站点连通性和凭证有效性。返回 data 按站点组织,每站含 ok(布尔)和 detail(诊断信息)。总体 success=true 不代表每站都可用,须读取各站 ok。不抛异常,自检失败也返回 success=true。
按明确选定的 ID 批量下载媒体文件。post_ids 为数组,单个 ID 也用数组。返回 data.downloaded 数组,每项含单作品的下载结果(post/directory/files/sidecar_path/errors/complete)。已有文件默认复用(overwrite=false),强制重新下载设 overwrite=true。部分失败时 success=false,error.type=partial_download,已完成文件保留。
按作品页面 URL 直接下载,无需先调用搜索。自动识别站点和 post_id,返回单作品下载结果(post/directory/files/sidecar_path/errors/complete)。已有文件默认复用(overwrite=false),强制重新下载设 overwrite=true。
获取单个作品的完整元数据(不下载媒体)。返回 data 为单个 Post 对象,含 site/id/url/title/page_count/tags/artist/rating/score 等。作品不存在时返回 error.type=not_found。
按关键词发现作品,只读取搜索与元数据,不下载媒体、不创建下载目录。适合用户尚未提供作品 ID/链接时调用。返回 data.posts(含 site/id/url/title/page_count)、count/page/limit。count=0 表示当前页无匹配结果,可调整关键词、过滤条件或页码。需要保存媒体时,把选定的 posts[].id 放入 media_hunter_download 的 post_ids 数组(单个 ID 也用数组)。
STDIO-only transport prevents remote accessibility and hosted MCP client integration. This is a hard architectural limitation.
Parameter descriptions are verbose (100-200+ chars each) and use Chinese, which may reduce LLM comprehension. Descriptions should be 50-100 chars in English for optimal LLM parsing.
Error handling lacks explicit recovery guidance. media_hunter_check returns success=true even on self-check failures, and error.type values (not_found, partial_download) are not documented with recovery paths.
media_hunter_check has empty input schema ({}), making it unclear what parameters it accepts or whether it requires configuration. Tool should document if it auto-detects credentials or requires explicit site selection.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 69 | 2025-06-18+ | v2 |
No dry-run or confirmation pattern for destructive operations (media_hunter_download, media_hunter_download_url). Agents cannot preview what will be downloaded before committing.