fix(cua-driver-rs)(windows): stop idle overlay CPU + orphan mcp child (#1808)

docs/content/docs/cua-driver/reference/changelog.mdx

@@ -18,6 +18,20 @@ The canonical, always-current source is the
 release body is auto-generated per release from the commits touching
 `libs/cua-driver/rust`, including SHA256 checksums and install instructions.
 
+## 0.5.6 (unreleased)
+
+- **Windows: fix idle CPU burn and orphaned `mcp` processes from the cursor
+  overlay** (#1808). The agent-cursor overlay render timer ran at ~125 Hz
+  unconditionally and re-composited a full virtual-screen pixmap (plus an
+  RGBA→BGRA copy and `UpdateLayeredWindow` blit) on every tick — so an idle
+  `cua-driver mcp` with no automation pinned 60–85% of a CPU core. The render
+  loop is now event-driven: it only composites and blits while a cursor is
+  actually animating or fading, and drops to a slow heartbeat (waking instantly
+  on the next command) once every cursor is static. The in-process `mcp` (stdio)
+  child now force-exits when the client disconnects (stdin EOF / pipe closed)
+  instead of leaving the overlay loop running as an orphan that accumulates CPU.
+  Use `--no-overlay` to disable the agent cursor entirely for headless runs.
+
 ## 0.5.4 (2026-06-15)
 
 - **macOS: fix 0x0 screenshots on macOS 15+ — the release bundle now ships the

libs/cua-driver/rust/crates/cua-driver/src/main.rs

@@ -714,8 +714,21 @@ async fn async_main() -> anyhow::Result<()> {
     let registry = Arc::new(build_registry(cursor_cfg));
     registry.init_self_weak();
     maybe_init_pip();
-    cua_driver_core::server::run(registry).await?;
-    Ok(())
+    let result = cua_driver_core::server::run(registry).await;
+    if let Err(e) = &result {
+        tracing::error!("MCP server error: {e}");
+    }
+
+    // The stdio MCP server loop has ended — the client disconnected (stdin
+    // EOF) or a fatal I/O error occurred. The cursor overlay runs on its own
+    // detached thread with an independent Win32 message loop (and we raised the
+    // multimedia timer resolution via `timeBeginPeriod`), so simply returning
+    // is not guaranteed to tear it down promptly: that thread is never joined
+    // and would otherwise keep its render loop alive as an orphan, accumulating
+    // CPU after the client is gone (issue #1808). Force a clean process exit so
+    // the overlay thread dies with us the moment the transport closes — mirrors
+    // the macOS `std::process::exit(0)` after `server::run`.
+    std::process::exit(if result.is_ok() { 0 } else { 1 });
 }
 
 // ── Registry builder (non-macOS) ──────────────────────────────────────────