# AirMCP MCP server

> Enables AI assistants to interact directly with macOS and Google Workspace apps like Mail, Calendar, Notes, and Drive using natural language.

- Repository: https://github.com/heznpc/iConnect
- Page: https://hostedmcp.org/servers/heznpc/iconnect
- Transport: stdio
- Stars: 17
- License: MIT
- Latest version: 0.1.1

## Install

### Claude Code

```sh
claude mcp add iconnect -- npx -y iconnect-mcp
```

### Cursor (`~/.cursor/mcp.json`)

```json
{
  "mcpServers": {
    "iconnect": {
      "command": "npx",
      "args": [
        "-y",
        "iconnect-mcp"
      ]
    }
  }
}
```

### VS Code (`.vscode/mcp.json`)

```json
{
  "servers": {
    "iconnect": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "iconnect-mcp"
      ]
    }
  }
}
```

### VS Code CLI

```sh
code --add-mcp '{"name":"iconnect","type":"stdio","command":"npx","args":["-y","iconnect-mcp"]}'
```

### Claude Desktop (`claude_desktop_config.json`)

```json
{
  "mcpServers": {
    "iconnect": {
      "command": "npx",
      "args": [
        "-y",
        "iconnect-mcp"
      ]
    }
  }
}
```

### Codex

```sh
codex mcp add iconnect -- npx -y iconnect-mcp
```

### Codex config.toml (`~/.codex/config.toml`)

```toml
[mcp_servers.iconnect]
command = "npx"
args = ["-y", "iconnect-mcp"]
```

### Gemini CLI

```sh
gemini mcp add -s user iconnect npx -- -y iconnect-mcp
```

### Gemini settings.json (`~/.gemini/settings.json`)

```json
{
  "mcpServers": {
    "iconnect": {
      "command": "npx",
      "args": [
        "-y",
        "iconnect-mcp"
      ]
    }
  }
}
```

## FAQ

### What apps can AirMCP control?

It provides tools for Apple apps including Notes, Mail, Calendar, Reminders, Messages, Photos, Safari, Finder, and Shortcuts, as well as Google Workspace apps like Gmail, Drive, and Sheets.

### Is my data being copied to a sandbox?

No, the server interacts with your actual local data rather than a copy or a sandbox.

### How does AirMCP handle destructive actions?

AirMCP uses a safety model where destructive calls are previewed before they run, and approval is required on a per-call basis.
