instruckt-tauri

Tauri v2 plugin for instruckt — visual feedback for AI coding agents. Provides the Rust backend, IPC commands, JSON file storage, and a standalone MCP server.

Users annotate elements in the webview, annotations are copied as structured markdown, and your AI agent can also read them via MCP.

Dev-only — the annotation UI and IPC shim are only injected in debug builds (cfg!(debug_assertions)). Zero bytes in production.

• Rust 1.70+
• Tauri v2
• A Vite-based frontend (React, Vue, Svelte)

# src-tauri/Cargo.toml
[dependencies]
tauri-plugin-instruckt = { git = "https://github.com/Naoray/instruckt-tauri.git", tag = "v0.1.0" }

// src-tauri/src/main.rs (or lib.rs)
fn main() {
    tauri::Builder::default()
        .plugin(tauri_plugin_instruckt::init())
        .run(tauri::generate_context!())
        .expect("error while running tauri application");
}

// src-tauri/capabilities/default.json
{
  "identifier": "default",
  "windows": ["main"],
  "permissions": [
    "core:default",
    "instruckt:default"
  ]
}

The Vite plugin provides the client-side UI. Set server: false since Tauri owns the backend:

// vite.config.ts
import instruckt from 'instruckt/vite'

export default defineConfig({
  plugins: [
    instruckt({
      server: false,
      mcp: true,
    }),
  ],
})

That's it. Run cargo tauri dev and the annotation toolbar appears in the bottom-right corner.

The plugin includes a standalone MCP server binary that AI agents connect to via stdio.

cargo install --git https://github.com/Naoray/instruckt-tauri.git --tag v0.1.0 --bin instruckt-mcp

Add to .mcp.json in your project root:

{
  "mcpServers": {
    "instruckt": {
      "command": "instruckt-mcp"