Turn the user’s focus into useful context.

const document = new DocxViewer(canvas, {
  enableTextSelection: true,
  enableElementSelection: true,
  onSelectionContextChange(context) {
    updateAskAiButton(context);
  },
});

const workbook = new XlsxViewer(container, {
  enableElementSelection: true,
  onSelectionContextChange(context) {
    updateAskAiButton(context);
  },
});

const slides = new PptxViewer(canvas, {
  enableTextSelection: true,
  enableElementSelection: true,
  onSelectionContextChange(context) {
    updateAskAiButton(context);
  },
});

The callbacks report semantic changes, not assistant actions. Your application decides whether the action is explain, summarize, compare, evaluate, or something else; the Viewer does not grow one API for each prompt.

There is no separate onClick option. For a raw browser click, add a native click listener to a stable application-owned host that contains the Viewer. This also observes DOCX/PPTX text-overlay clicks, whose target is a sibling of the canvas. For the resulting read-only focus, use onSelectionContextChange. This keeps one semantic selection notification instead of a second callback for the same interaction.

Open a menu for the right-click target.

const viewer = new XlsxViewer(container, {
  enableElementSelection: true,
  onContextMenu: async ({ originalEvent, getContext }) => {
    // Suppress the browser menu before the first await.
    originalEvent.preventDefault();
    const { clientX, clientY } = originalEvent;
    const context = await getContext();
    openContextMenu({ clientX, clientY, context });
  },
});

originalEvent is the actual browser event, delivered synchronously so preventDefault() still controls the native menu. getContext() is explicitly asynchronous because a DOCX or PPTX object hit can require a worker query; it starts the lookup on its first call and returns the same memoized Promise if called again. Omit onContextMenu and the Viewer installs no listener. The event itself is never serialized or sent through MCP.