Swift Integration

CuaDriverCore and CuaDriverServer are available as Swift library products directly from the trycua/cua repository. Use this when you want to embed accessibility automation, window capture, or MCP tool handling into your own macOS Swift app or package — without shelling out to the cua-driver CLI.

Add the dependency

In your Package.swift:

dependencies: [
    .package(
        url: "https://github.com/trycua/cua.git",
        from: "cua-driver-v0.1.0"
    ),
],

Then add the products you need to your target:

targets: [
    .target(
        name: "MyApp",
        dependencies: [
            // Accessibility, input synthesis, window capture, recording.
            // No external dependencies — system frameworks only.
            .product(name: "CuaDriverCore", package: "cua"),

            // MCP tool handlers and daemon layer (adds swift-sdk dependency).
            // Use this if you want to expose cua-driver tools over MCP.
            .product(name: "CuaDriverServer", package: "cua"),
        ]
    ),
]

CuaDriverCore depends only on system frameworks (AppKit, CoreGraphics, ScreenCaptureKit, etc.). CuaDriverServer adds modelcontextprotocol/swift-sdk for the MCP protocol types.

What's in each product

CuaDriverCore

Low-level macOS automation primitives:

Module area	What it gives you
`AppStateEngine`	AX tree snapshots with element-index caching
`AppLauncher`	Launch apps without focus steal
`AXInput`	Perform AX actions, read attributes
`MouseInput` / `KeyboardInput`	Synthesize CGEvents
`WindowCapture`	Per-window screenshots via ScreenCaptureKit
`RecordingSession`	Trajectory recording and replay
`AgentCursor`	Animated overlay cursor
`BrowserJS` / `ElectronJS`	Execute JS in browser tabs via Apple Events or CDP
`AXPageReader`	Extract text / query DOM from WKWebView AX trees

CuaDriverServer

Ready-made MCP tool handlers built on CuaDriverCore:

import CuaDriverServer

let registry = ToolRegistry.default
// registry.handlers["click"], ["get_window_state"], ["page"], etc.

Embed the full tool set in your own MCP server:

import MCP
import CuaDriverServer

let server = MCPServer(tools: ToolRegistry.default.allTools) { name, args in
    try await ToolRegistry.default.call(name, arguments: args)
}

Minimum requirements

macOS 14 (Sonoma) or later
Swift 6.0+

TCC permissions

Calling AX or screen-capture APIs from an embedded library still requires the host app to hold the relevant TCC grants:

Accessibility — for any AXUIElement work (snapshots, clicks, AX actions).
Screen Recording — for WindowCapture screenshots.

Grant these under System Settings → Privacy & Security against your app's bundle identifier, the same as for the standalone CuaDriver.app.

Was this page helpful?