How to Add MCP Tools to LM Studio on Mac

How LMCP sets up LM Studio — Mac: Automatically (installer and menu) — Restart the app · Windows: From the app's own MCP settings

When the stack stays local

LM Studio runs open LLMs locally on your Mac. LMCP is a local MCP server that gives an AI access to your real Mac apps. With an on-device model and exclusively local integrations, model inference and LMCP tool execution stay on your Mac without cloud AI or API keys. Connected services such as Teams or OneDrive can synchronize or receive requested data under their own terms, and any cloud-backed AI provider or integration you select can receive content requested for that task. Ask LM Studio to read your email, check your calendar, or search your Messages, and it acts on the real apps:

• “Summarize my unread emails”
• “What’s on my calendar today?”
• “Find the iMessage where Ana sent the address”

Step 1: Install LMCP

Download and open the installer (macOS 13+, Apple Silicon & Intel):

curl -fsSL 'https://local-mcp.com/install?ref=guide-lm-studio-mac' | bash

Or download the app. LMCP starts in your menu bar.

Step 2: Add LMCP to LM Studio

LM Studio loads MCP servers from ~/.lmstudio/mcp.json. Open it inside the app: in the right sidebar switch to the Program tab, click Install → Edit mcp.json, and add LMCP under mcpServers:

{
  "mcpServers": {
    "local-mcp": {
      "command": "npx",
      "args": ["-y", "local-mcp@latest"]
    }
  }
}

Save the file — LM Studio loads the server automatically. (LM Studio uses npx to launch it, so you need Node.js installed.)

Step 3: Enable and try it

Make sure the local-mcp server is toggled on, load a tool-capable model, start a chat, and ask:

List my email accounts

LM Studio calls the tool on your Mac and returns the result. With an on-device model and only local integrations, that workflow stays on the Mac; connected services can still synchronize or receive data under their own terms.

Troubleshooting

The server won’t start — make sure the Local MCP app is running (menu-bar icon) and that npx is on your PATH (install Node.js).

The model doesn’t call tools — pick a model that supports tool/function calling; not every local model does.