This server provides MCP tools to search and browse the local Windows filesystem via the Everything index, with advanced query modes and metadata filtering.
Perform fast file and folder searches using Everything's native query syntax (e.g., ext:pdf, size:>1mb, dm:today).
Search by exact filename (including extension) with no wildcard interpretation.
Match filenames containing all/any specified keywords, with optional exclusions.
Use wildcard patterns (* and ?) for filename matching.
Use regular expressions for advanced filename searches.
Fuzzy search with subsequence, substring, or unordered keyword matching.
Search using Chinese pinyin (full, initials, mixed) with optional heteronym support.
Filter by file category (document, image, video, audio, code, archive).
List directory contents, optionally recursively.
Get counts of matching files/folders without retrieving details.
Check server status including ES version and IPC connectivity.
Apply additional filters by size range, modification date range, and path, and paginate results with offset/limit.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@everything-mcp find the latest annual report PDF on my desktop"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
English | 简体中文
通过 MCP 调用 Windows 本机 Everything 索引，查找文件和目录。提供 12 个搜索及辅助工具，支持路径、扩展名、大小和修改日期筛选，返回完整路径及文件元数据。
当前版本仅支持在 Windows 上运行，依赖本机 Everything 和 ES；macOS、Linux 及其他 Unix 类系统暂不支持原生运行。
需安装 Everything、ES 和 uv（提供 uvx），Python 版本要求为 3.11+。通过 PyPI 安装，无需 Git 或克隆仓库。
从 voidtools 下载页 安装 Everything，并保持运行。
在同一页面下载 Everything Command-line Interface（ES），解压到固定目录。
将 es.exe 所在目录加入 Windows 用户或系统 PATH，然后重启终端和 MCP 客户端。
Everything 与 ES 是不同组件。如果已安装全局 ES，且客户端能从 PATH 找到它，无需设置 EVERYTHING_ES_PATH。
验证安装：
es -versionRelated MCP server: everything-search-mcp
使用 uvx evryth-mcp 从 PyPI 自动安装并启动，无需提前执行 pip install。
适用于使用 mcpServers JSON 格式的客户端：
{
"mcpServers": {
"everything-es": {
"command": "uvx",
"args": ["evryth-mcp"]
}
}
}连接后调用 everything_status，确认返回 connected: true。
如果 ES 已在客户端 PATH 中且使用默认 Everything 实例，上面的配置即可，无需添加环境变量。
需要指定 ES 路径或命名实例时，在 everything-es 配置中添加 env：
{
"mcpServers": {
"everything-es": {
"command": "uvx",
"args": ["evryth-mcp"],
"env": {
"EVERYTHING_ES_PATH": "D:\\Tools\\Everything\\es.exe",
"EVERYTHING_INSTANCE": "Work"
}
}
}
}EVERYTHING_ES_PATH：替换为实际 es.exe 的绝对路径，不是 Everything.exe 或目录；客户端能从 PATH 找到 ES 时可删除此项。
EVERYTHING_INSTANCE：填写已运行的 Everything 命名实例名称，示例为 Work；使用默认实例时删除此项。它不会自动创建或启动实例。
这两项可独立使用；如果都不需要，删除整个 env。
工具 | 功能 |
everything_search | Everything 原生查询、排序和分页 |
everything_search_exact | 完整文件名匹配 |
everything_search_keywords | 全部/任一关键词及排除词 |
everything_search_wildcard | * 和 ? 通配符 |
everything_search_regex | 正则表达式 |
everything_search_fuzzy | 连续、非连续及无序关键词匹配 |
everything_search_pinyin | 全拼、首字母、中文混输及多音字 |
everything_search_by_type | 文档、图片、视频、音频、代码、压缩包 |
everything_search_typo | 文件名字符相似度匹配 |
everything_list_directory | 直属或递归目录索引项 |
everything_count | 匹配数量 |
everything_status | 版本与 IPC 状态 |
例如，让客户端“查找包含合同和2025、排除草稿的 PDF”，或“按拼音 niandubaogao 搜索年度报告”。具体参数由客户端工具描述提供。
默认搜索 Everything 全部索引，不额外限制目录；未索引的文件不会出现。
没有专门的文件内容搜索或文件修改工具。
拼音和容错按候选分页匹配，单页无结果不代表全盘不存在；容错分数仅在当前页排序。
IPC 不可达时，检查 Everything 是否运行，以及客户端会话和沙箱权限。
终端能执行 ES 而客户端找不到时，先完全重启客户端，使其继承最新 PATH。
Count matching indexed files/folders without retrieving their paths.
Inputs: query (required), kind, path, regex, +2 more.
Browse indexed children (not a filesystem listing). recursive=False returns direct children; missing index entries are not included.
Inputs: path (required), kind, limit, offset, +2 more.
Search the local Everything index. query supports Everything syntax (ext:pdf, size:>1mb, dm:today). No path means all indexed locations. Returns paths, size bytes and modification times. Pagination reflects a live, changing index. Does not read file contents.
Inputs: query (required), kind, path, sort, +7 more.
Search a documented extension category. Explicit filters.extensions overrides the category's default extension list.
Inputs: category (required), path, text, limit, +2 more.
Match the entire literal basename, including extension. No wildcard interpretation.
Inputs: name (required), kind, path, limit, +2 more.
Filename matching: subsequence (约指南 matches 约会指南), substring, or unordered keywords. Literal characters are escaped; use search_typo for spelling errors.
Inputs: text (required), kind, mode, path, +3 more.
Match all/any literal filename keywords and exclude literal words; words containing spaces remain phrases.
Inputs: keywords (required), kind, path, limit, +4 more.
Match full pinyin, initials or mixed Chinese/pinyin against a bounded ES candidate page. heteronym enables alternative character readings. Continue next_candidate_offset until candidates_exhausted; not an exhaustive global result before that.
Inputs: text (required), mode, path, limit, +5 more.
Search filenames using ES regular expressions, e.g. ^报告.*.pdf$.
Inputs: pattern (required), kind, path, limit, +2 more.
Rank a bounded page of filenames by SequenceMatcher similarity (0..100). Not semantic search. All matches on the page returned; ranking is page-local. Continue next_candidate_offset for remaining candidates.
Inputs: text (required), path, filters, threshold, +3 more.
Match full basename using * (zero or more characters) and ? (one character). Other characters are literal.
Inputs: pattern (required), kind, path, limit, +2 more.
Check ES version and Everything IPC connectivity. No changes to the service.
Inputs: no named parameters.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 12 tools
Multiple tools search filenames with different matching modes (regex, fuzzy, pinyin, exact, keywords, wildcard, typo), which creates real overlap in purpose. The descriptions do a good job differentiating them, but an agent must read carefully to select the correct variant.
All tools follow a consistent everything_verb[_modifier] snake_case pattern, with the search variants uniformly named everything_search_*. This makes the tool family easy to recognize and predict.
Twelve tools is well within the ideal range and each tool has a distinct matching mode or core operation. The count feels justified for a search-oriented MCP server covering the Everything index.
The surface covers the core Everything operations: general search, specialized filename matching, counting, status checks, and directory browsing. No obvious dead ends or missing critical capabilities for working with the local Everything index.
Read and search folders on your own computer with explicit consent. Files are never uploaded.
Search the web, images, videos, news, and local businesses with robust filters, freshness controls…
Search prompts, CLAUDE.md/AGENTS.md templates, Cursor rules, Agent Skills and MCP servers.
Search, browse, and read your Dropbox files. Find documents by name or content, list folders, and…