loonghao/auroraview/qt-to-auroraview-migration

Qt → AuroraView migration skill

This skill turns an existing Qt-based Python project (typically one that uses QWebEngineView, a custom QMainWindow, or a DCC tool dialog with embedded HTML) into an AuroraView project with:

• A native Rust-powered WebView (~5 MB vs ~120 MB Qt WebEngine)
• Unified create_webview() entry point (standalone / Qt parent / HWND)
• Bidirectional Python ↔ JavaScript bridge (webview.emit / @webview.on)
• Full MCP control out of the box via dcc-mcp-core's gateway — every

AuroraView window registers itself into the gateway and exposes a CDP bridge, so agents can click, fill, screenshot, and call Python APIs over standard MCP without the user writing any Python glue

When to use this skill

Use it when you see any of these signals in the source project:

1. from PySide6.QtWebEngineWidgets import QWebEngineView (or PySide2 / PyQt5 / PyQt6)
2. QWebChannel + registerObject for JS↔Python IPC
3. A DCC-embedded Qt dialog that loads an HTML/React/Vue frontend
4. A QMainWindow whose central widget is basically a browser
5. The user says "convert/migrate/port this Qt tool to AuroraView" or

"I want to control this Qt app with MCP"

Decision tree

Is there a QWidget parent (Maya/Houdini/Nuke/3ds Max)?
├── Yes → use QtWebView path (create_webview(parent=qwidget, ...))
└── No
    ├── Is there a native HWND (Unreal, C++ host)?
    │   └── Yes → use HWND path (create_webview(parent=hwnd, ...))
    └── No → standalone (create_webview(url=..., ...))

All three paths return the same WebView API; only the parent argument differs.

Migration playbook

Perform these steps in order. Each has a concrete before/after snippet so you can grep-and-replace safely.

Step 1 — Dependencies

Replace Qt WebEngine with AuroraView. Keep qtpy / a Qt binding only if the tool still needs Qt widgets around the webview (toolbars, dock panels, DCC integration).

# pyproject.toml — remove
# "PySide6-WebEngine"
# "PyQtWebEngine"
# "auroraview-mcp"              # the standalone Python MCP pkg was retired; see #364
 
# pyproject.toml — add
"auroraview>=0.4.19"            # core
"auroraview[qt]>=0.4.19"        # if embedding inside a Qt host
# MCP control now comes from dcc-mcp-core's gateway + auroraview-cli; no extra
# Python dep is needed on the project itself. Install the tooling separately:
#   pip install dcc-mcp-core    # gateway + skill runtime (PyPI)
#   auroraview-cli skills install --target cursor    # wire agent tools

Step 2 — Replace QWebEngineView

# BEFORE
from PySide6.QtWebEngineWidgets import QWebEngineView
from PySide6.QtCore import QUrl
view = QWebEngineView(parent)
view.setUrl(QUrl("http://localhost:3000"))
view.show()

# AFTER
from auroraview import create_webview
webview = create_webview(parent=parent, url="http://localhost:3000")
webview.show()

When there is no Qt parent, pass nothing:

webview = create_webview(url="http://localhost:3000", title="My Tool")
webview.show()

Step 3 — Replace QWebChannel IPC

AuroraView ships a first-class bridge — drop QWebChannel, registerObject, and the custom JS adapter entirely.

# BEFORE — Qt WebChannel
from PySide6.QtWebChannel import QWebChannel
class Bridge(QObject):
    @Slot(dict, result=dict)
    def export_scene(self, data):
        ...
channel = QWebChannel(view.page())
channel.registerObject("bridge", Bridge())
view.page().setWebChannel(channel)

# AFTER — AuroraView
@webview.on("export_scene")
def handle_export(data):
    return {"ok": True, "path": data["path"]}
# Python → JS
webview.emit("scene_updated", {"frame": 120})

On the JS side, replace the qtwebchannel.js adapter with the AuroraView SDK:

// BEFORE
new QWebChannel(qt.webChannelTransport, (channel) => {
  channel.objects.bridge.export_scene({ path });
});
// AFTER
import { av } from "@auroraview/sdk";
await av.call("export_scene", { path });
av.on("scene_updated", (data) => console.log(data));

Step 4 — Replace QMainWindow browser shells

If the project is a custom browser (tabs, URL bar, history), keep the Qt chrome but swap every QWebEngineView for AuroraView. The @examples/qt_browser.py file in this repo is the before; here is the AFTER pattern:

from auroraview import create_webview
class BrowserTab(QWidget):
    def __init__(self, url: str, parent=None):
        super().__init__(parent)
        layout = QVBoxLayout(self)
        layout.setContentsMargins(0, 0, 0, 0)
        self.webview = create_webview(parent=self, url=url)
        layout.addWidget(self.webview.widget())  # expose the native widget

Note: for native tab switching/reloading, call webview.load_url(...), webview.reload(), webview.back(), webview.forward().

Step 5 — DCC integration (Maya / Houdini / Nuke / 3ds Max)

Thread-dispatcher backends are already bundled. The migration is literally:

# BEFORE
ptr = omui.MQtUtil.mainWindow()
parent = wrapInstance(int(ptr), QWidget)
view = QWebEngineView(parent)
view.setUrl(QUrl("http://tool"))
# AFTER
from auroraview import create_webview
webview = create_webview(parent=parent, url="http://tool")  # autoloads qt backend

For non-Qt DCCs (Unreal) pass the HWND as an int:

webview = create_webview(parent=unreal_hwnd, url="http://tool")
hwnd = webview.get_hwnd()  # needed by the host engine

Step 6 — Enable MCP control (the "for free" part)

Once the app is on AuroraView, agents drive it over MCP with zero extra code on the project side. AuroraView windows register themselves into the shared dcc-mcp-core gateway and expose their CDP endpoint automatically; agents discover them and call tools through the gateway.

# One-time setup on the developer machine:
pip install dcc-mcp-core                         # gateway + MCP runtime
auroraview-cli skills install --target cursor    # or --target claude / all
 
# Then simply run the migrated app:
python my_tool.py     # AuroraView registers itself with the gateway on start

Agents see tools exposed through dcc-mcp-core's progressive discovery — tools/list stays small because the gateway hides detail-level tools behind load_skill. For AuroraView the shipped skills are:

See Epic #364 for the rollout status — the Rust adapter crate (crates/auroraview-mcp) and the skills listed above are tracked under issues #365–#368.

To expose custom Python methods to agents, decorate them and bind via auroraview.api:

from auroraview import create_webview, api
class ToolAPI:
    def rename_selected(self, prefix: str = "av_") -> dict:
        ...
    def export_scene(self, path: str) -> dict:
        ...
webview = create_webview(parent=parent, url="http://tool")
api.bind(webview, ToolAPI())
# Now agents can call: mcp.call_api("rename_selected", {"prefix": "hero_"})

One-shot migration checklist

Use this checklist when applying the skill to a repo. Tick every item.

• [ ] Grep for QWebEngineView / QWebEnginePage / QWebChannel — zero hits
• [ ] Grep for PySide*-WebEngine / PyQtWebEngine in deps — zero hits
• [ ] from auroraview import create_webview replaces WebEngine imports
• [ ] @webview.on(...) / webview.emit(...) replaces QWebChannel slots
• [ ] JS side uses @auroraview/sdk instead of qwebchannel.js
• [ ] Installed dcc-mcp-core + ran auroraview-cli skills install
• [ ] Verified with the agent: calling list_windows from

auroraview-diagnostics returns at least the migrated app's window

• [ ] All DCC entry points use create_webview(parent=...) (never QWebEngineView)
• [ ] Sample/smoke test: agent load_skill("auroraview-devtools") then

capture_screenshot returns a PNG

• [ ] README migration note added

References (read these on demand)

• python/auroraview/api.py — unified create_webview() contract
• python/auroraview/__init__.py — full public API surface
• python/auroraview/integration/qt/ — Qt embedding internals
• examples/qt_browser.py — canonical BEFORE example (pure Qt WebEngine)
• examples/maya_qt_echo_demo.py — canonical AFTER example (Qt host + AuroraView + API bridge)
• packages/auroraview-sdk/ — JS-side bridge (@auroraview/sdk)
• crates/auroraview-mcp/ — Rust adapter crate that plugs AuroraView into

dcc-mcp-core's gateway (tracks Epic #364; part of the new MCP story)

• `dcc-mcp-core` upstream — gateway,

skill runtime, and progressive tool discovery; MCP tools ship as SKILL.md packages delivered via auroraview-cli skills install

Troubleshooting