Tauri MCP

Tauri MCP Plugin Skill

Enables AI agents (Codex, 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 Codex 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())
))

Codex MCP Configuration

Add to .Codex/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:

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 Codex settings (.Codex/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 (Codex / 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

Troubleshooting

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