tauri-mcp

name	tauri-mcp
description	Setup, configure, and manage the Tauri MCP plugin for AI agent debugging of the Foundry desktop app. Use when adding MCP support, configuring the MCP server, troubleshooting MCP connections, or when the user invokes /tauri-mcp.
version	1.0.0
tags	["tauri","mcp","desktop","debugging","ai-agent"]

Tauri MCP Plugin Skill

Enables AI agents (Claude Code, Cursor) to interact with and debug the Foundry Tauri desktop app through screenshots, DOM access, input simulation, and more via the Model Context Protocol.

When to Use

• Setting up the MCP plugin in the Tauri app
• Configuring Claude Code or Cursor to connect to the running desktop app
• Troubleshooting MCP connection issues (socket, TCP, auth)
• Adding or modifying MCP tools exposed to AI agents
• Debugging the desktop app via MCP inspector

Project Context

The Foundry desktop app lives at apps/desktop/:

• Rust backend: apps/desktop/src-tauri/ (Cargo workspace)
• Frontend: apps/desktop/src/ (Vite + React 19)
• Cargo.toml already includes tauri-plugin-mcp as a git dependency

Quick Reference

Dependencies

Rust (already in Cargo.toml):

tauri-plugin-mcp = { git = "https://github.com/P3GLEG/tauri-plugin-mcp" }

JS guest bindings (install if needed):

bun add tauri-plugin-mcp -D --filter @foundry/desktop

MCP server CLI (for agent connection):

npx tauri-plugin-mcp-server

Rust Plugin Registration

Register in apps/desktop/src-tauri/src/lib.rs — dev builds only:

#[cfg(debug_assertions)]
{
    builder = builder.plugin(tauri_plugin_mcp::init_with_config(
        tauri_plugin_mcp::PluginConfig::new("foundry-desktop".to_string())
            .start_socket_server(true)
            .socket_path("/tmp/foundry-mcp.sock")
    ));
}

TCP mode (alternative):

.plugin(tauri_plugin_mcp::init_with_config(
    tauri_plugin_mcp::PluginConfig::new("foundry-desktop".to_string())
        .tcp_localhost(4000)
        .auth_token("your-secret-token".to_string())
))

Claude Code MCP Configuration

Add to .claude/settings.json or project MCP config:

IPC mode (default, recommended):

{
  "mcpServers": {
    "tauri-mcp": {
      "command": "npx",
      "args": ["tauri-plugin-mcp-server"]
    }
  }
}

Custom socket path:

{
  "mcpServers": {
    "tauri-mcp": {
      "command": "npx",
      "args": ["tauri-plugin-mcp-server"],
      "env": {
        "TAURI_MCP_IPC_PATH": "/tmp/foundry-mcp.sock"
      }
    }
  }
}

TCP mode:

{
  "mcpServers": {
    "tauri-mcp": {
      "command": "npx",
      "args": ["tauri-plugin-mcp-server"],
      "env": {
        "TAURI_MCP_CONNECTION_TYPE": "tcp",
        "TAURI_MCP_TCP_HOST": "127.0.0.1",
        "TAURI_MCP_TCP_PORT": "4000"
      }
    }
  }
}

Available MCP Tools

Once connected, AI agents have access to 10 tools:

Tool	Purpose	Key Parameters
take_screenshot	Capture window screenshot	Full image saved to disk, inline thumbnail
query_page	DOM inspection	Modes: map, html, state, find_element, app_info
click	Click elements	By coordinates (x,y) or selector (ref, id, class, tag, text)
type_text	Fill text fields	fields array for bulk fill, supports contentEditable
mouse_action	Mouse operations	Hover, scroll (direction/amount), drag
navigate	Webview navigation	goto URL, back/forward (with delta), reload
execute_js	Run JavaScript	Arbitrary JS in webview context, returns result
manage_storage	Storage operations	localStorage get/set/remove/clear/keys, cookies
manage_window	Window control	List/focus/minimize/maximize/close/position/size/fullscreen, zoom, devtools
wait_for	Wait for conditions	Text appearing/disappearing, element visible/hidden/attached/detached

See docs/tools-reference.md for detailed parameter documentation.

Setup Workflow

1. Register the Plugin (Rust)

Modify apps/desktop/src-tauri/src/lib.rs to register the plugin in cfg(debug_assertions) block. The builder chain needs to include the MCP plugin before .run().

2. Install JS Guest Bindings (if using JS API)

cd apps/desktop && bun add tauri-plugin-mcp -D

3. Configure Agent MCP Server

Add the MCP server config to your Claude Code settings (.claude/settings.json) or Cursor settings.

4. Start the App and Verify

# Start the desktop app in dev mode
cd apps/desktop && bun run tauri dev

# In another terminal, test with MCP inspector
npx @modelcontextprotocol/inspector npx tauri-plugin-mcp-server

Architecture

AI Agent (Claude Code / Cursor)
  ↓ stdio
MCP Server (tauri-plugin-mcp-server)
  ↓ IPC socket or TCP
Tauri Plugin (Rust: socket server, command routing, native input)
  ↓ Tauri events (correlation IDs)
Guest JS (DOM interaction, element resolution, form filling)
  ↓
DOM / Application

Platform Notes

Platform	Input Method	Screenshot Method
macOS	Native NSEvent injection (no Accessibility perms needed)	Native capture
Windows	JS fallback (~80% coverage, isTrusted=false)	Native capture
Linux	JS fallback (~80% coverage)	xcap library

Troubleshooting

Issue	Fix
Connection refused	Verify the Tauri app is running with socket server enabled
Socket not found	Check socket path matches between Rust config and MCP server env
Auth failed (TCP)	Ensure auth_token matches in Rust config and MCP server env
Stale socket	Plugin auto-cleans stale sockets on startup; restart the app
No screenshot	Ensure the window is visible and not minimized
Tools not responding	Check that #[cfg(debug_assertions)] is active (dev build only)

Diagnostic command:

npx @modelcontextprotocol/inspector npx tauri-plugin-mcp-server

Security

• Plugin should only be registered in debug builds (#[cfg(debug_assertions)])
• TCP mode requires auth token for non-loopback connections
• Auth tokens use constant-time comparison
• Token files are created with 0o600 permissions and cleaned up on shutdown