everything-semantic-mcp
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| EVERYTHING_INSTANCE | No | Named Everything instance (e.g. "1.5a") | |
| EVERYTHING_SEMANTIC_DATA | No | Semantic index storage location (default %LOCALAPPDATA%\everything-semantic-mcp) | |
| EVERYTHING_SEMANTIC_EXTS | No | Document extensions to include in the semantic index, separated by semicolons | |
| EVERYTHING_SEMANTIC_MODEL | No | Embedding model (default minishlab/potion-multilingual-128M) | |
| EVERYTHING_SEMANTIC_SCOPE | No | Restrict the semantic index target using Everything syntax (e.g. path:"D:\Work") | |
| EVERYTHING_SEMANTIC_EXCLUDE | No | Additional excluded paths, separated by semicolons | |
| EVERYTHING_SEMANTIC_IDLE_SEC | No | Seconds before the model is released from memory when unused (default 900) |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": false
} |
| prompts | {
"listChanged": false
} |
| resources | {
"subscribe": false,
"listChanged": false
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| everything_searchB | Search file and folder names across every drive and network folder Everything indexes. query: Everything syntax, e.g. 'ext:xlsx budget dm:last30days', 'path:"D:\Projects" *.pptx',
'folder: invoices', '"exact name.txt"'. An empty query matches everything (use with filters).
limit: 1-500 rows; |
| everything_countA | Count matches for up to 20 queries in one call (no rows returned), e.g. compare 'ext:pdf' vs 'ext:docx' under a folder, or check whether a file exists anywhere. |
| everything_recentA | Most recently modified files, newest first. path_filter: restrict to a folder or share. extensions: ';'-separated, e.g. 'xlsx;csv'. within: today | yesterday | last7days | last30days | thisweek | thismonth | 2026-10. |
| everything_largestA | Largest files (optionally under a folder or share), for disk-usage triage. min_size: '100mb', '1gb'. |
| everything_healthA | Index health and configuration self-check: version, build (x86/x64), database state, indexed volumes and folders, database age, and settings that cause slow indexing, RAM bloat or freezes (a drive indexed twice, change monitoring on network folders, foreground rescans, rescans at every startup, an index never saved, the 32-bit build). Each issue has a fix. |
| everything_maintainA | Index maintenance. save_db: persist the in-memory index (fast restart). rescan_folders: refresh folder indexes (network shares) now. rebuild: full re-index (slow; only if results look stale). |
| everything_semantic_searchA | Meaning-based search over folder names and document file names (multilingual, fully local). Use when the user describes WHAT they want rather than a literal name: 'last quarter's sales deck', '세금 계산서 모음', 'onboarding material'. Scores: >0.6 strong, 0.4-0.6 related. path_prefix restricts to a drive or share. Folders end with ''; list their contents with everything_search('parent:""'). |
| everything_semantic_reindexA | (Re)build the local semantic index from the live Everything database. Runs in the background unless wait=true (~3-7 min per million names on a laptop CPU); poll everything_semantic_status. |
| everything_semantic_statusC | Semantic index metadata (count, built_at, model), its location, and any running or failed build. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 9 tools
Each tool has a recognizable purpose: literal search vs semantic search is clearly split, and count/recent/largest/health/maintain/reindex/status all target distinct operations. The only mild overlap is that everything_recent and everything_largest are effectively presets of everything_search with a sort argument, but the descriptions frame them distinctly enough that misselection is unlikely.
Every tool uses the same everything_ prefix followed by a clear action descriptor (search, count, recent, largest, health, maintain, and the semantic_ family). The verb_noun-ish pattern is consistent across the whole set, including the namespaced semantic group.
Nine tools is well within the ideal 3-15 range for a focused file-search server. Each earns its place: two search modes, three aggregation helpers, two index-health tools, and two semantic-index lifecycle tools.
The surface covers literal and semantic discovery, aggregation (count/recent/largest), and index health/maintenance/reindex lifecycle well, so core workflows have no dead ends. Minor gaps remain, such as opening/revealing the file found or exporting results, which agents would have to work around.