diff --git a/db/migrations/20260707012459_add_media_file_title_sort_covering_index.sql b/db/migrations/20260707012459_add_media_file_title_sort_covering_index.sql
new file mode 100644
index 000000000..18666eef9
--- /dev/null
+++ b/db/migrations/20260707012459_add_media_file_title_sort_covering_index.sql
@@ -0,0 +1,22 @@
+-- +goose Up
+-- +goose StatementBegin
+
+-- Covering index for the title-sorted, library-scoped song listing:
+--   WHERE missing = ? AND library_id = ? ORDER BY order_title LIMIT n OFFSET m
+-- (Jellyfin clients page through the whole library this way; non-admin native and
+-- Subsonic song lists produce the same shape.)
+--
+-- Without it, SQLite walks media_file_order_title and must fetch the table row for
+-- every *skipped* entry just to evaluate the WHERE, so a deep page costs offset+limit
+-- random row reads (seconds on cold spinning disks). With the filter columns in the
+-- index the skip is index-only. `id` is included because the annotation/bookmark
+-- LEFT JOINs run per candidate row and need the join key; without it each skipped
+-- entry still triggers a row fetch.
+create index if not exists media_file_missing_library_order_title
+	on media_file(missing, library_id, order_title, id);
+-- +goose StatementEnd
+
+-- +goose Down
+-- +goose StatementBegin
+drop index if exists media_file_missing_library_order_title;
+-- +goose StatementEnd