
Xiaoyao Search
This is a local file search engine that runs on your machine and connects to Claude through MCP. It gives you five search modes: semantic search powered by BGE-M3 vectors, traditional full-text with Whoosh, hybrid that combines both, image similarity via CN-CLIP, and voice search with FasterWhisper transcription. You need to run the backend server yourself at localhost:8000 and build indexes before searching. The semantic search is genuinely useful for natural language queries like "machine learning optimization docs" while full-text handles exact matches like code snippets. It returns relevance scores and file previews in JSON. If you work with large local document collections and want more than grep, this gives you proper semantic understanding without sending files to the cloud.
- GitHub stars
- 1.1K
- last updated
- May 22, 2026
Install
$ npx -y skills add dtsola/xiaoyaosearch --skill xiaoyao-search --agent claude-codeInstalls into .claude/skills of the current project.
Files
When to use
当需要搜索本地文件时使用此 Skill:
- 通过自然语言描述搜索本地文档
- 查找包含特定关键词的文件
- 用图片搜索相似的本地图片
- 用语音转文字后搜索文件
- 需要综合语义和全文搜索结果
Capability
小遥搜索 MCP 服务器提供以下 5 个搜索工具:
1. semantic_search(语义搜索)
- 用途:基于 BGE-M3 向量模型的语义搜索
- 适用:自然语言查询,如"关于机器学习的文档"
- 参数:
query: string - 搜索关键词(1-500字符)limit: int - 返回结果数(1-100,默认20)threshold: float - 相似度阈值(0.0-1.0,默认0.5)file_types: string[] - 文件类型过滤(可选)
2. fulltext_search(全文搜索)
- 用途:基于 Whoosh 的精确关键词匹配
- 适用:查找特定术语、代码片段
- 参数:
query: string - 搜索关键词(1-500字符)limit: int - 返回结果数(1-100,默认20)file_types: string[] - 文件类型过滤(可选)
3. hybrid_search(混合搜索)
- 用途:结合语义和全文搜索,RRF 算法融合结果
- 适用:需要全面搜索结果的场景
- 参数:
query: string - 搜索关键词(1-500字符)limit: int - 返回结果数(1-100,默认20)threshold: float - 相似度阈值(0.0-1.0,默认0.5)file_types: string[] - 文件类型过滤(可选)
4. image_search(图像搜索)
- 用途:基于 CN-CLIP 的图像相似度搜索
- 适用:用图片查找相似的本地文件
- 参数:
image_path: string - 图片绝对路径limit: int - 返回结果数(1-100,默认20)threshold: float - 相似度阈值(0.0-1.0,默认0.5)
Sponsors
Details
- Repository
- github.com/dtsola/xiaoyaosearch
- Category
- AI & Agent Building
- First seen
- Jun 11, 2026