Semlink

中文

An Obsidian plugin that vectorizes your Vault notes and exposes semantic search via MCP (Model Context Protocol).

Let Claude Desktop, Claude Code, Cursor and other AI tools directly search and read your Obsidian notes.

Features

• Semantic Search: Query your Vault notes in natural language, return the most relevant results based on vector similarity
• Real-time Indexing: Automatically detect file changes and update the vector index
• MCP Server: Expose search capabilities via HTTP protocol for AI clients

MCP Tools

How It Works

Obsidian Vault Notes
        │
        ▼
   Text Chunking
        │
        ▼
  SiliconFlow API (BGE-M3)  ──→  Vector Embedding
        │
        ▼
  Local Storage (SQLite + Binary)
        │
        ▼
  MCP HTTP Server (:3001)
        │
        ▼
  Claude / Cursor / Other AI Clients

Installation

Option 1: Build from Source (Developers)

# Clone the repo
git clone https://github.com/ouou365/Semlink.git
cd semlink

# Install dependencies
npm install

# Build
npm run build

# Copy the entire directory to your Obsidian Vault plugins folder
# e.g. MyVault/.obsidian/plugins/semlink/

Option 2: Direct Download

Download main.js, manifest.json, styles.css, sql-wasm.wasm from the Release page and place them in:

YourVault/.obsidian/plugins/semlink/

Enable the Plugin

1. Open Obsidian → Settings → Community plugins
2. Find Semlink and enable it

Configuration

After enabling the plugin, go to Settings → Semlink:

Getting an API Key

1. Register at siliconflow.cn
2. Create an API Key in the console
3. Enter the key in plugin settings

Connect AI Clients

After configuration, client configs are auto-generated at the bottom of the plugin settings page.

Claude Desktop / Cursor

Add the following JSON to your MCP configuration file:

{
  "mcpServers": {
    "semlink": {
      "type": "http",
      "url": "http://127.0.0.1:3001/mcp"
    }
  }
}

With MCP access key:

{
  "mcpServers": {
    "semlink": {
      "type": "http",
      "url": "http://127.0.0.1:3001/mcp",
      "headers": {
        "Authorization": "Bearer your-key"
      }
    }
  }
}

Claude Code

Run in terminal:

# Without key
claude mcp add --transport http semlink http://127.0.0.1:3001/mcp

# With key
claude mcp add --transport http semlink http://127.0.0.1:3001/mcp --header "Authorization: Bearer your-key"

Usage

First Indexing

1. Configure your API Key
2. Open command palette (Ctrl/Cmd + P), search for "Semlink: Start Index"
3. Click Semlink in the status bar to view indexing progress

Daily Use

• The plugin automatically watches for file changes and incrementally updates the index
• Ask questions in your AI client to search your notes

Commands

Data Storage

All data is stored locally only:

These files are located in the plugin directory and are never uploaded to any server.

Embedding Models

Tech Stack

• Embedding: SiliconFlow API (BGE-M3, 1024-dim)
• Storage: sql.js (SQLite WASM) + binary vector file
• Search: Brute-force cosine similarity
• Protocol: MCP over HTTP (JSON-RPC 2.0)

Development

# Dev mode (watch and auto-rebuild)
npm run dev

# Production build
npm run build

Notes

• Full indexing consumes SiliconFlow API credits; watch your usage for large Vaults
• Vector search runs in local memory; 100K+ notes may use significant memory
• MCP server listens on 127.0.0.1 by default, accessible only from localhost

License

MIT