Quickstart

Korely is a memory layer for your AI agents. It runs as an MCP server, exposes 7 tools (search, read, list_notes, list_folders, get_related, save_note, create_task) and stores everything as plain Markdown in a vault on your disk.

This guide gets you from zero to working integration in about five minutes. You'll need any MCP-capable client (Claude Code, Cursor, ChatGPT desktop, n8n, Continue, or Goose).

What you'll have at the end: Korely installed, the vault directory created at ~/Korely, and your AI client able to call any of the 7 MCP tools against your vault.

1. Install Korely

Korely ships as a desktop app for macOS, Windows, and Linux. The desktop app also embeds the local MCP server — no separate install needed.

Korely is currently in private beta — join the early access waitlist and we'll send the installer and onboarding link as soon as your slot opens up. On first launch you'll be asked to pick a vault location. The default ~/Korely works for everyone; pick something else only if you already have an Obsidian or markdown notes folder you want to point at.

Existing Obsidian vault? Korely reads any folder of .md files. Point it at your Obsidian vault and the knowledge graph builds itself on first scan.

2. Locate the MCP server binary

After install, the MCP server lives inside the app bundle. The exact path depends on your OS:

OSPath
macOS /Applications/Korely.app/Contents/Resources/mcp-server
Windows %LOCALAPPDATA%\Korely\resources\mcp-server.exe
Linux (AppImage) ~/.korely/mcp-server

Or use the npm package if you prefer that wiring (works on every platform): 

npx -y @korely/mcp-server

The npx route is what every integration guide on this site uses by default — easier to copy-paste, easier to keep up to date.

3. Wire it to your AI client

Every MCP-capable client has a config file or settings panel where you declare MCP servers. The exact location varies:

The minimum config is always the same shape:

{
  "mcpServers": {
    "korely": {
      "command": "npx",
      "args": ["-y", "@korely/mcp-server"],
      "env": {
        "KORELY_VAULT_ROOT": "~/Korely"
      }
    }
  }
}

KORELY_VAULT_ROOT tells the server where your vault lives. The default is ~/Korely (created by the desktop app on first launch). Set it to a custom path if you point at an Obsidian vault or moved the directory.

4. Verify the connection

Restart your AI client after editing the MCP config. The 7 Korely tools should appear in the tool picker (the exact UI is client-specific; in Claude Code it's /tools; in Cursor it's the wrench icon in the chat panel).

The fastest smoke test — ask the assistant:

"Use the korely_search tool to find notes about 'test'."

If you have any note in your vault containing "test", it should come back in seconds with the matching snippet. If the vault is empty, the tool returns an empty array — that's a success too, it means the server is talking to your vault.

Tool picker empty? Most often the client hasn't picked up the new MCP config. Quit the client fully (Cmd-Q, not just close window) and relaunch. If still nothing, check the client's MCP log — Claude Code logs to ~/.config/claude-code/logs/mcp.log; Cursor surfaces errors in Settings → MCP.

5. Next steps

You're connected. Where to go from here:

  • Wire a specific client: pick the integration guide that matches your tool of choice and follow the step-by-step. Each guide includes copy-paste config snippets and the common pitfalls.
  • Understand the 7 tools: read the MCP tools reference to see what your agent can do and how to prompt for each one.
  • Ship it to production: when you need multi-tenant cloud MCP (OAuth, rate limits, audit log), see pricing. The Developer tier wraps the same 7 tools with cloud auth and R2-backed vault sync.

Stuck somewhere? Email hello@korely.ai with your client name and what's failing. I read every message and respond within a day. We're early — your feedback shapes what gets fixed first.