Summary

The speed comes from Everything, and so does the dependency.

This is a thin, well-shaped adapter over an index that already exists on the machine, which is exactly why results are instant where a recursive directory walk would take minutes. The flip side is that nothing works until Everything is installed and its HTTP server switched on — that toggle, not the npm install, is the step people miss. Two tools is the right number here: structured search for the query you can express, natural search for the one you can only describe.

What it is

An MCP server in front of Everything Search Engine's HTTP interface. Because Everything keeps a live index of the filesystem, results come back immediately no matter how large the drive — the server exposes that index as two tools, one structured and one that accepts a plain-language description.

What you get

`search` for precise queries — `query` plus `scope`, with `caseSensitive`, `wholeWord`, `regex` and `path` flags, `maxResults`, and sorting through `sortBy` and `ascending`
`natural_search` for the other direction — a `naturalQuery` such as 'find all video files larger than 1GB' and a result limit
Sorting by name, path, size or modification date, so 'the largest' or 'the most recent' is a parameter rather than a follow-up
Search restricted to a directory by setting the scope, or left across the whole drive
Human-readable file sizes and formatted dates in the results

Requirements

Windows, because Everything Search Engine is Windows-only. Install Everything, then enable its HTTP Server under Tools > Options > HTTP Server on port 8011 and restart it — the MCP server talks to that interface. Also Node.js 16+. Install the npm package `everything-search-mcp-server` globally, or build from source, then point the client at the `everything-search-mcp-server` command.