repository·main·Indexed 42 days ago
https://github.com/hypothesi/mcp-server-tauriA Model Context Protocol (MCP) server that enables AI assistants like Claude, Cursor, and Windsurf to build, test, and debug Tauri v2 applications. It provides tools for UI automation, IPC monitoring, log streaming, and webview interaction via the tauri-plugin-mcp-bridge Rust plugin.
The @hypothesi/tauri-mcp-cli package is a bundled skill designed for AI coding agents to automate or test Tauri v2 desktop and mobile applications. It provides a comprehensive surface covering session management, UI interaction, inspection, IPC debugging, and mobile workflows.
Covered Domains:
Prompts are pre-built templates that appear as slash commands in MCP-compatible AI assistants (like Claude Desktop or Cursor). Unlike tools, which the AI calls automatically, prompts are user-controlled workflows.
webview_screenshot)./fix-webview-errors).Use prompts when you need a guided debugging session, a testing flow, or a complex task that requires multiple tools working together in a specific sequence.
Note: Some MCP clients (like Windsurf) may not support slash commands yet. If they don't work, you can manually instruct the AI to follow the prompt's logic.
The Test App is specifically configured to support the MCP server's native automation through several key mechanisms:
9223-9322 range for UI automation.packages/tauri-plugin-mcp-bridge/. It provides IPC monitoring and direct command execution.When performing UI automation or element inspection, choose the tool based on who initiates the selection workflow:
webview_select_element: Use this when the agent initiates the picker. This is best when the AI wants to actively ask the user to select something specific.webview_get_pointed_element: Use this when the user initiates the selection. This is best when the user has already used the Alt+Shift+Click shortcut and tells the agent to look at what they pointed at.The CLI distinguishes between the background process and the active driver connection:
daemon stop: Kills the background process entirely.driver-session stop: Only ends the current connection to the application; the daemon may still be running.Best Practice: Because driver-session start might report success even if the app is not live, always verify the connection immediately after starting with:
driver-session status --jsonVerify that the output contains "connected": true.
The MCP server can connect to multiple Tauri applications simultaneously. Each application must run on a unique port.
Key behaviors:
port number or bundleId via the appIdentifier parameter.// Connect to first app
await driver_session({ action: "start", port: 9223 })
// Connect to second app
await driver_session({ action: "start", port: 9224 })
// Check status - shows both apps with default indicator
await driver_session({ action: "status" })
// Use default app (the most recent one, e.g., port 9224)
await webview_screenshot()
// Target a specific app by its port
await webview_screenshot({ appIdentifier: 9223 })
// Stop a specific app session
await driver_session({ action: "stop", appIdentifier: 9223 })
// Stop all active app sessions
await driver_session({ action: "stop" })The MCP Bridge acts as a communication layer between a Node.js-based MCP Server and a Tauri application.
The system uses a three-tier architecture to bridge AI assistants with Tauri applications:
Communication Flow:
AI Assistant $\rightarrow$ MCP Protocol (stdio) $\rightarrow$ MCP Server $\rightarrow$ WebSocket (port 9223) $\rightarrow$ Tauri Plugin $\rightarrow$ Webview (JavaScript)
In the context of the Tauri MCP server, Prompts are pre-built templates that guide your AI assistant through multi-step, structured workflows. While Tools are individual actions (e.g., taking a single screenshot), Prompts combine multiple tools into a cohesive experience.
| Aspect | Tools | Prompts |
|---|---|---|
| Control | AI-controlled | User-controlled |
| Scope | Single action | Multi-step workflow |
| Invocation | AI decides when to use | You type a slash command |
| Example | tauri_webview_screenshot | /fix-webview-errors |
In MCP clients like Claude Desktop, Cursor, and VS Code, these appear as slash commands (e.g., /command-name).
The MCP Server for Tauri acts as a bridge between your AI assistant and your running Tauri application.
Communication Flow:
mcp-bridge plugin installed in your app.The /fix-webview-errors prompt initiates a complete debugging workflow to find and fix JavaScript errors in your Tauri app's webview.
driver_session.read_logs.withGlobalTauri must be enabled in tauri.conf.json.Example usage:
/fix-webview-errorsTo enable AI-powered development for your Tauri app, you must first register the MCP server with your AI assistant (e.g., Claude Code, Cursor, Windsurf, VS Code).
Run the following command to automatically add the server:
npx -y @a1st/aix add mcp tauri --command 'npx @hypothesi/tauri-mcp-server' --userFor a direct terminal workflow, install and use the CLI:
npm install -g @hypothesi/tauri-mcp-cli
tauri-mcp driver-session start --port 9223Add the following to your MCP configuration file:
{
"mcpServers": {
"tauri": {
"command": "npx",
"args": ["-y", "@hypothesi/tauri-mcp-server"]
}
}
}Config file locations:
Cursor Settings → MCP → New MCP Serversettings.json under mcp.serversNote: Restart your AI assistant after making these changes.