diff --git a/script/rubocop-diff.sh b/script/rubocop-diff.sh
index 32dbef1ca8..d1dca94e04 100755
--- a/script/rubocop-diff.sh
+++ b/script/rubocop-diff.sh
@@ -1,14 +1,41 @@
 #!/bin/bash
 #
 # While you are developing, you can call this script to check all
-# changed files. And then you can also tell Git to check before every
+# changed files. You can auto-correct them if you wish:
+#
+#   ./script/rubocop-diff.sh -a
+#
+# And then you can also tell Git to check before every
 # commit by adding this line to your `.git/hooks/pre-commit` file:
 #
 #   ./script/rubocop-diff.sh --cached || exit 1
 #
 
-rubocop="`dirname $0`/../bin/rubocop"
-cached="$1" # may be empty
+# If you prefer to use faster boot times of the Rubocop Server then you can
+# define your way of calling rubocop:
+#
+#   export RUBOCOP_BIN="rubocop --server"
+#
+# Or locked to the bundled version (needs update after `bundle update`):
+#
+#   export RUBOCOP_BIN="`bundle show rubocop`/exe/rubocop --server"
+#
+# I don't know how to set that automatically though.
+#
+# But I observed some performance improvement:
+#
+# * Using default binstup with spring: boot: 6.2s, repeat: 0.4s, 0.4s, ...
+# * Using rubocop server binstub without bundler: boot: 2s, repeat: 1s, 0.3s, ...
+# * Using rubocop executable directly: boot: 2.1s, repeat: 1s, 0.16s, ...
+#
+# The default binstub is still the safest, always using the bundled version.
+: ${RUBOCOP_BIN="`dirname $0`/../bin/rubocop"}
+
+if [ "$1" = "--cached" ]; then
+  cached="$1"
+else
+  rubocop_opt="$1"
+fi
 
 if git diff $cached --diff-filter=ACMR HEAD --quiet; then
 	# nothing changed
@@ -17,7 +44,9 @@ fi
 
 exec git diff $cached --name-only --relative --diff-filter=ACMR HEAD |\
 	xargs \
-	$rubocop --force-exclusion \
-                 --fail-level A \
-                 --format simple \
-                 --parallel --cache true
+	$RUBOCOP_BIN \
+          --force-exclusion \
+          --fail-level A \
+          --format simple \
+          --parallel --cache true \
+	  "$rubocop_opt"