# obsidian-brain [Health: Active]

**Category:** 🧠 Knowledge & Memory  
**Repository:** https://github.com/sweir1/obsidian-brain  
**GitHub Stars:** 13  
**npm Downloads (last month):** 23439  
**Views:** 4  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/obsidian-brain

## Description
Obsidian MCP server: semantic search, knowledge graph, and vault editing. No plugin required.

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

```json
"mcpServers": {
  "obsidian-brain": {
    "command": "npx",
    "args": ["-y","cache"],
    "env": {
      "VAULT_PATH": ""
    }
  }
}
```

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

## Documentation

## What obsidian-brain MCP server does

The obsidian-brain MCP server connects an MCP client to an Obsidian vault or any directory containing Markdown files. It supports three main workflows: finding and reading content, examining relationships between notes, and changing files. The server can create, edit, link, move, and delete notes, with an edit-preview operation available before applying a change.

Its retrieval tools include semantic search, note listing, and note reading. Graph operations can find related notes, calculate paths between two notes, identify themes, and rank notes using graph structure. The project lists 18 tools in total, including indexing and index-status operations. Optional live-editor tools are available only when the companion Obsidian plugin is installed.

## How it works

The process runs locally as a Node application and exchanges JSON-RPC messages with the MCP client through stdio. It reads vault files from disk, parses their contents, and stores searchable data in SQLite. The index contains note nodes, link edges, full-text data, and vector embeddings.

Content is split into heading-aware chunks, with code and LaTeX blocks preserved. Search can use hybrid retrieval, combining semantic ranking with SQLite FTS5 BM25 results through Reciprocal Rank Fusion. When a note changes, the server writes to the vault immediately and incrementally re-indexes the affected file. A first launch indexes the vault and downloads an embedding model of about 34 MB; later launches avoid that initial work.

## Setup and configuration

The obsidian-brain MCP server requires Node.js 20 or newer and a configured `VAULT_PATH`. A typical client configuration starts the package with `npx -y obsidian-brain@latest server` and passes the vault location as an environment variable. The supplied example targets Claude Desktop, while the project also documents setup for other MCP clients.

No Obsidian installation is needed for the core server, and Obsidian may be closed while files are read or edited. The repository also provides a macOS installer that configures dependencies and the Claude Desktop connection. Users who prefer manual setup can add the MCP server entry themselves.

## Tools and capabilities

Available capabilities include:

- Search, list, and read Markdown notes.
- Find connections and paths across the vault graph.
- Detect themes and rank notes with graph analytics.
- Create, edit, preview, link, move, and delete notes.
- Rebuild the index and inspect indexing status.
- Query the active note, Dataview, and Bases through the optional companion plugin.

The optional plugin exposes live Obsidian runtime state through a localhost HTTP endpoint and requires Obsidian to be running. Server and plugin releases should use the same major and minor versions.

## Limitations and notes

The obsidian-brain MCP server is local and uses the configured filesystem path; it does not provide a hosted endpoint. The core feature set does not include live editor state, Dataview results, or Bases rows without the companion plugin. Initial indexing can delay tool availability for 30–60 seconds.

A missing `VAULT_PATH` prevents startup. Native SQLite dependencies can also fail when built for a different Node ABI, requiring a rebuild. The documentation notes that stale `npx` caches may cause an older release to load. Vault content is intended to remain on the local machine according to the project description.

_Full upstream README: https://allmcps.com/mcp/obsidian-brain/readme_