# vault-knowledge [Health: Active]

**Category:** 🔒 Security  
**Repository:** https://github.com/seanwinslow28/sw-mcp-vault-knowledge  
**GitHub Stars:** 0  
**npm Downloads (last month):** 241  
**Views:** 3  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/vault-knowledge

## Description
Read-only MCP over a vault's typed knowledge graph: concept search, contradictions, article fetch.

## Tools
Capabilities this server exposes over MCP:

- **search_concepts** — Semantically search the vault's concept, connection, and Q&A articles for a natural-language query. Returns ranked {slug, title, similarity_score, excerpt, last_modified}. Read-only.
- **find_contradictions** — Traverse the typed reasoning graph for active 'contradicts' edges — the sharpest of six relation types. scope='all' or 'recent_30d'. Returns {from_slug, to_slug, confidence, surfaced_at, source_run_id}. Read-only.
- **get_article** — Fetch one article by slug with frontmatter, body, and resolved inbound/outbound wikilinks. Reads only files under the configured knowledge allowlist. Read-only.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "vault-knowledge": {
    "command": "npx",
    "args": ["-y","@swins/vault-knowledge-mcp"],
    "env": {
      "VAULT_DB": "",
      "VAULT_ROOT": ""
    }
  }
}
```

**Requires environment variables:** `VAULT_DB`, `VAULT_ROOT` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What vault-knowledge MCP server does

The vault-knowledge MCP server makes selected content from an Obsidian vault available to MCP clients as a read-only typed knowledge graph. It focuses on three operations: semantic concept search, contradiction discovery, and article retrieval. The server is intended for agents that need to inspect an existing body of notes rather than edit, reorganize, or synchronize the vault.

The graph models relationships between concept and connection articles, including a `contradicts` relation. This lets an agent identify disagreements recorded in the notes instead of only finding text that matches a query. Article retrieval includes frontmatter, the article body, and resolved wikilinks.

## How it works

The vault-knowledge MCP server runs locally through stdio, so an MCP client launches it as a subprocess. A synthetic espresso vault is included for an immediate test without configuring a personal data source. To use an existing vault, the process receives the path to an index database and the vault root through environment variables.

`search_concepts` ranks relevant concept and connection articles semantically. On its first execution, the server downloads the `nomic-embed-text-v1.5` model. `find_contradictions` reads active `contradicts` edges from the reasoning graph and supports either the full graph or a recent-30-day scope. `get_article` loads one article by slug and resolves its internal links.

## Setup and configuration

The documented installation uses Node.js and the npm package `@swins/vault-knowledge-mcp`:

```json
{
  "mcpServers": {
    "vault-knowledge": {
      "command": "npx",
      "args": ["-y", "@swins/vault-knowledge-mcp"]
    }
  }
}
```

Node.js 22.5 or later is required because the server uses the built-in `node:sqlite` module. The default configuration starts with the included synthetic vault. For a personal vault, set `VAULT_DB` to the index database path and `VAULT_ROOT` to the vault directory.

## Tools and capabilities

- `search_concepts(query, limit=5)` returns semantically ranked concept and connection articles.
- `find_contradictions(scope='all'|'recent_30d')` returns active contradiction edges, either across the graph or from the recent period.
- `get_article(slug)` returns an article's frontmatter, body, and resolved wikilinks.

## Limitations and notes

The vault-knowledge MCP server is read-only and does not provide tools for editing notes or changing the graph. It has no authentication flow and does not require a cloud service. Only content under `knowledge/concepts`, `knowledge/connections`, and `knowledge/qa` is readable. This scope is configured for the server rather than supplied as a request parameter, preventing a client from widening access through a tool call.

The embedding model download occurs on first concept-search use, so that initial operation requires network access to obtain the model. Later searches can use the downloaded model. The README does not document named client-specific integrations; it describes the tools as callable by MCP clients generally.

_Full upstream README: https://allmcps.com/mcp/vault-knowledge/readme_