Tauri

You can connect to a MCP server that bridges your Tauri application with AI agents, enabling you to capture screenshots, inspect the DOM, and simulate user input from a client. This makes it possible for agents to understand and interact with your app in real-time through the Model Context Protocol (MCP).

How to use

To use the MCP server, run your Tauri app with the MCP integration and connect an MCP client to the Unix domain socket the server exposes. The client can request screenshots, read DOM content, find elements, and simulate clicks, typing, dragging, or hovering. You can also execute JavaScript inside the webview context from the client.

How to install

Prerequisites: you need Rust and Cargo installed to build the MCP server component. You also need Node.js if you plan to run the included example MCP server using Node.

Steps to prepare and run the MCP server with the example setup:

# Cargo.toml
[dependencies]
taori-mcp = { git = "https://github.com/DonsWayo/tauri-mcp", optional = true }

// src-tauri/src/lib.rs
#[cfg(feature = "mcp-e2e")]
app.plugin(tauri_plugin_mcp::init())?;

Configuration and usage notes

The MCP server opens a Unix domain socket at /tmp/<app-name>-mcp.sock. MCP clients connect to this socket to send requests for actions like capturing a screenshot, reading the DOM, or performing input events.

Example MCP client configuration

You can wire up an MCP client to run a local MCP server process and point it at the socket path. The following example shows a stdio-based setup that uses Node to host the MCP server and connects to the socket.

{
  "mcpServers": {
    "nucel_e2e": {
      "type": "stdio",
      "command": "node",
      "args": ["path/to/mcp-server/build/index.js", "/tmp/your-app-mcp.sock"]
    }
  }
}

Additional notes

Screenshot capture on macOS, Linux, and Windows is supported. The macOS setup uses a system-friendly screenshot method that avoids requiring Screen Recording permissions for the Tauri app itself. On Linux, window capture is performed via X11 with multi-pass window matching by app name and title.

Available tools

screenshot

Capture the current app window as an image to provide visual context to the MCP client.

dom_read

Read the full HTML of the active webview window to enable content understanding and element selection.

element_interact

Simulate user input actions such as click, type, drag, and hover on DOM elements.

js_eval

Execute arbitrary JavaScript in the webview context to manipulate or query the page.

element_position

Locate elements by id, class, tag, or text content to enable targeted interactions.