diff --git a/core/musicbrainz_search.py b/core/musicbrainz_search.py
index ff05be13..ebd5184c 100644
--- a/core/musicbrainz_search.py
+++ b/core/musicbrainz_search.py
@@ -638,13 +638,34 @@ class MusicBrainzSearchClient:
             logger.warning(f"MusicBrainz track search failed: {e}")
             return []
 
-    def _search_tracks_text(self, track_name: str, artist_name: Optional[str], limit: int) -> List[Track]:
-        """Fallback text-search path for structured/fuzzy track queries."""
+    def _search_tracks_text(self, track_name: str, artist_name: Optional[str], limit: int,
+                            strict: bool = True, min_score: Optional[int] = None) -> List[Track]:
+        """Fallback text-search path for structured/fuzzy track queries.
+
+        `strict=True` (default) keeps the field-scoped Lucene phrase match —
+        precise enough for enrichment-style flows where the inputs are
+        already known-clean. `strict=False` switches to a bare-query
+        MB lookup that hits alias/sortname indexes with diacritic folding —
+        needed for user-facing fuzzy surfaces (Fix popup cascade) where
+        recall beats precision because the user picks from the result list.
+        Mirrors the same toggle already on `search_recording` in
+        `core/musicbrainz_client.py`.
+
+        `min_score` defaults to `self._MIN_SCORE` (80) — sized for the
+        enhanced search tab where unfiltered MB results are noisy. Pass
+        a lower value (or 0) when a downstream stage like
+        `core.metadata.relevance.rerank_tracks` will re-sort by artist
+        match — MB's free-text score heavily favours title-text matches
+        ("Army of Me (Bjork)" cover by HIRS Collective scores 100,
+        Björk's canonical "Army of Me" scores 28) so a high floor drops
+        the right answer.
+        """
         try:
-            results = self._client.search_recording(track_name, artist_name=artist_name, limit=limit)
-            # Score filter matches the artist/album logic — cuts garbage
-            # title collisions from unrelated recordings.
-            results = [r for r in results if (r.get('score', 0) or 0) >= self._MIN_SCORE]
+            results = self._client.search_recording(
+                track_name, artist_name=artist_name, limit=limit, strict=strict
+            )
+            threshold = self._MIN_SCORE if min_score is None else min_score
+            results = [r for r in results if (r.get('score', 0) or 0) >= threshold]
 
             tracks = []
             for r in results:
@@ -656,6 +677,30 @@ class MusicBrainzSearchClient:
             logger.warning(f"MusicBrainz track search failed: {e}")
             return []
 
+    def search_tracks_with_artist(self, track: str, artist: str,
+                                  limit: int = 10) -> List[Track]:
+        """Search MB tracks with track + artist passed as separate fields.
+
+        Powers the Fix-popup metadata cascade (`GET /api/musicbrainz/search_tracks`)
+        and any future surface where the caller already has the title/artist
+        split and wants the fuzzy-recall MB lookup without going through
+        `search_tracks`'s structured-query dispatch (`Artist - Track`
+        splitting, bare-name artist-first browse).
+
+        Uses bare-query mode (`strict=False`) — diacritic-folded, hits
+        alias/sortname indexes, no `AND`-clause that kills recall when
+        either side mis-matches. Score floor lowered to 20 (vs the search
+        tab's 80) so MB recordings whose title doesn't literally contain
+        the artist name still enter the candidate pool — the endpoint's
+        `rerank_tracks` pass then sorts by artist-match relevance. Without
+        this, queries like `Army of Me` + `Bjork` only surface covers
+        (score 73-100) and miss Björk's canonical recording (score 28).
+        """
+        if not track and not artist:
+            return []
+        return self._search_tracks_text(track, artist or None, limit,
+                                        strict=False, min_score=20)
+
     def _pick_representative_release(self, releases: List[Dict[str, Any]]) -> Optional[Dict[str, Any]]:
         """Pick the best release out of a release-group's editions.
 
diff --git a/tests/metadata/test_musicbrainz_search.py b/tests/metadata/test_musicbrainz_search.py
index 350a52a4..0a97a89b 100644
--- a/tests/metadata/test_musicbrainz_search.py
+++ b/tests/metadata/test_musicbrainz_search.py
@@ -887,3 +887,135 @@ def test_get_recording_flat_swallows_client_errors():
     client._client.get_recording.side_effect = RuntimeError('boom')
 
     assert client.get_recording_flat('rec-err') is None
+
+
+# ---------------------------------------------------------------------------
+# search_tracks_with_artist — Fix-popup cascade adapter
+# ---------------------------------------------------------------------------
+
+def test_search_tracks_with_artist_uses_bare_query_mode():
+    """The Fix-popup cascade needs MB's bare-query mode so diacritics and
+    bracketed suffixes don't kill recall. The adapter must pass strict=False
+    through to the underlying search_recording call."""
+    client = MusicBrainzSearchClient()
+    client._client = MagicMock()
+    client._client.search_recording.return_value = [
+        {'id': 'rec-1', 'title': 'Army of Me', 'score': 95,
+         'releases': [{'id': 'rel-1', 'title': 'Post', 'date': '1995'}],
+         'artist-credit': [{'name': 'Björk'}]},
+    ]
+
+    tracks = client.search_tracks_with_artist('Army of Me', 'Björk', limit=10)
+
+    # strict=False is the critical bit — fuzzy recall, not phrase precision
+    client._client.search_recording.assert_called_once_with(
+        'Army of Me', artist_name='Björk', limit=10, strict=False
+    )
+    assert len(tracks) == 1
+    assert tracks[0].name == 'Army of Me'
+    assert 'Björk' in tracks[0].artists
+
+
+def test_search_tracks_with_artist_handles_missing_artist():
+    """Track-only query (no artist) still works — empty string becomes
+    None, and the underlying client searches recordings without an
+    artist filter."""
+    client = MusicBrainzSearchClient()
+    client._client = MagicMock()
+    client._client.search_recording.return_value = [
+        {'id': 'rec-1', 'title': 'Some Song', 'score': 90,
+         'releases': [], 'artist-credit': [{'name': 'Unknown'}]},
+    ]
+
+    client.search_tracks_with_artist('Some Song', '', limit=5)
+
+    # Empty artist → None passed to the client so MB drops the AND clause
+    client._client.search_recording.assert_called_once_with(
+        'Some Song', artist_name=None, limit=5, strict=False
+    )
+
+
+def test_search_tracks_with_artist_empty_returns_empty_list():
+    """No track and no artist → return [] without hitting the network."""
+    client = MusicBrainzSearchClient()
+    client._client = MagicMock()
+
+    assert client.search_tracks_with_artist('', '', limit=10) == []
+    client._client.search_recording.assert_not_called()
+
+
+def test_search_tracks_with_artist_keeps_low_score_for_rerank():
+    """Cascade path uses a low score floor (20) so MB recordings whose
+    title doesn't literally contain the artist name still enter the
+    candidate pool — the endpoint's rerank pass surfaces them by
+    artist-match relevance. Real example: "Army of Me" + "Bjork" — the
+    canonical Björk recording scores 28 in MB (title doesn't contain
+    "Bjork"), while title-collision covers like "Army of Me (Bjork)"
+    score 73-100. Strict 80 floor drops the right answer."""
+    client = MusicBrainzSearchClient()
+    client._client = MagicMock()
+    client._client.search_recording.return_value = [
+        {'id': 'rec-cover', 'title': 'Army of Me (Bjork)', 'score': 100,
+         'releases': [], 'artist-credit': [{'name': 'HIRS Collective'}]},
+        {'id': 'rec-canonical', 'title': 'Army of Me', 'score': 28,
+         'releases': [], 'artist-credit': [{'name': 'Björk'}]},
+        {'id': 'rec-noise', 'title': 'Bjork', 'score': 5,
+         'releases': [], 'artist-credit': [{'name': 'Random'}]},
+    ]
+
+    tracks = client.search_tracks_with_artist('Army of Me', 'Bjork', limit=50)
+
+    ids = [t.id for t in tracks]
+    # Score=28 canonical Björk recording is kept — the endpoint's rerank
+    # will surface it by artist match.
+    assert 'rec-canonical' in ids
+    assert 'rec-cover' in ids
+    # Score=5 is below the 20 floor — true garbage still filtered out.
+    assert 'rec-noise' not in ids
+
+
+def test_search_tracks_text_keeps_min_score_default_80_for_enhanced_search():
+    """The enhanced search tab path keeps the historical 80 floor because
+    it has no downstream rerank — unfiltered MB results would be noisy
+    for free-text user search."""
+    client = MusicBrainzSearchClient()
+    client._client = MagicMock()
+    client._client.search_recording.return_value = [
+        {'id': 'rec-good', 'title': 'Good', 'score': 95,
+         'releases': [], 'artist-credit': [{'name': 'A'}]},
+        {'id': 'rec-mid', 'title': 'Mid', 'score': 40,
+         'releases': [], 'artist-credit': [{'name': 'A'}]},
+    ]
+
+    # No min_score → defaults to _MIN_SCORE (80)
+    tracks = client._search_tracks_text('Good', 'A', limit=10)
+
+    titles = [t.name for t in tracks]
+    assert 'Good' in titles
+    assert 'Mid' not in titles
+
+
+def test_search_tracks_text_strict_param_default_true():
+    """Default strict=True preserves the historical behaviour of the
+    structured-query text-search fallback path — important so the
+    enrichment-style `search_tracks('Artist - Track')` flow stays on
+    field-scoped Lucene phrase matching as before."""
+    client = MusicBrainzSearchClient()
+    client._client = MagicMock()
+    client._client.search_recording.return_value = []
+
+    client._search_tracks_text('Track', 'Artist', limit=10)
+
+    client._client.search_recording.assert_called_once_with(
+        'Track', artist_name='Artist', limit=10, strict=True
+    )
+
+
+def test_search_tracks_with_artist_swallows_client_errors():
+    """MB client raising must not crash the endpoint — return [] so the
+    Fix-popup cascade falls through to the next source."""
+    client = MusicBrainzSearchClient()
+    client._client = MagicMock()
+    client._client.search_recording.side_effect = RuntimeError('network down')
+
+    assert client.search_tracks_with_artist('Track', 'Artist', limit=10) == []
diff --git a/web_server.py b/web_server.py
index 347e938c..b735c7c7 100644
--- a/web_server.py
+++ b/web_server.py
@@ -19190,6 +19190,69 @@ def search_deezer_tracks():
         return jsonify({"error": str(e)}), 500
 
 
+@app.route('/api/musicbrainz/search_tracks', methods=['GET'])
+def search_musicbrainz_tracks():
+    """Search for tracks on MusicBrainz — used by the Discovery Fix popup
+    cascade and any future surface that needs track-level MB search in the
+    Fix-popup track shape.
+
+    Mirrors the spotify / itunes / deezer search_tracks endpoints exactly:
+    accepts `track` + `artist` (or legacy `query`) plus `limit`, returns
+    `{tracks: [{id, name, artists, album, duration_ms, image_url, source}]}`.
+
+    Uses MB's bare-query mode for max recall (diacritic-folded,
+    alias/sortname indexed) — same rationale as the manual MBID-paste
+    endpoint shipped earlier. The Fix popup is a user-facing fuzzy search
+    where the user picks from the result list, so recall beats precision.
+    """
+    try:
+        track_q = request.args.get('track', '').strip()
+        artist_q = request.args.get('artist', '').strip()
+        legacy_query = request.args.get('query', '').strip()
+        limit = int(request.args.get('limit', 20))
+
+        if not (track_q or artist_q or legacy_query):
+            return jsonify({"error": "Query parameter is required"}), 400
+
+        from core.musicbrainz_search import MusicBrainzSearchClient
+        mb_search = MusicBrainzSearchClient()
+
+        if track_q or artist_q:
+            tracks = mb_search.search_tracks_with_artist(
+                track_q or legacy_query, artist_q, limit=limit
+            )
+        else:
+            # Legacy single-string query — let MB's structured-query
+            # dispatch decide artist-first browse vs text search.
+            tracks = mb_search.search_tracks(legacy_query, limit=limit)
+
+        # Local rerank — same helper Deezer / iTunes use. Penalises
+        # cover / karaoke / tribute patterns + boosts exact-artist match.
+        if track_q or artist_q:
+            from core.metadata.relevance import rerank_tracks
+            tracks = rerank_tracks(
+                tracks,
+                expected_title=track_q,
+                expected_artist=artist_q,
+            )
+
+        tracks_dict = [{
+            'id': t.id,
+            'name': t.name,
+            'artists': t.artists,
+            'album': t.album,
+            'duration_ms': t.duration_ms,
+            'image_url': t.image_url,
+            'source': 'musicbrainz',
+        } for t in tracks]
+
+        return jsonify({'tracks': tracks_dict})
+
+    except Exception as e:
+        logger.error(f"Error searching MusicBrainz tracks: {e}")
+        return jsonify({"error": str(e)}), 500
+
+
 @app.route('/api/itunes/album/<album_id>', methods=['GET'])
 def get_itunes_album_tracks(album_id):
     """Fetches full track details for a specific iTunes album."""
diff --git a/webui/static/helper.js b/webui/static/helper.js
index 0b581183..ca5f330d 100644
--- a/webui/static/helper.js
+++ b/webui/static/helper.js
@@ -3422,6 +3422,7 @@ const WHATS_NEW = {
         { title: 'Fix: MusicBrainz manual search missing results', desc: 'the Fix popup and manual library service search were using strict Lucene phrase-match queries against the `recording` / `release` / `artist` fields — diacritics ("Bjork" vs canonical "Björk"), bracketed suffixes like "(Live)", and any AND-clause mismatch all killed recall. switched user-facing manual lookups to bare queries that hit MB\'s alias / sortname indexes with diacritic folding. enrichment workers stay strict for precision.' },
         { title: 'Fix: MusicBrainz album clicks 404ing in enhanced search', desc: 'every click on a MusicBrainz album result was silently 404-ing — the /release fetch was passing `cover-art-archive` as an `inc` param, which MB rejects with 400 (that field is returned on every release response by default, no include needed). dropped the bad include; album detail now loads correctly.' },
         { title: 'Fix popup: paste a MusicBrainz URL or MBID to match directly', desc: 'new escape hatch on the Fix Track Match modal (the 🔧 Fix button on mirrored / YouTube / Tidal / Deezer / Beatport / ListenBrainz / Spotify-public discovery rows). when fuzzy search keeps ranking the wrong recording among many same-title versions, paste the MusicBrainz recording URL like `https://musicbrainz.org/recording/<uuid>` or the bare UUID into the new field and hit "Look up". skips all fuzzy logic, resolves straight to that record, and runs it through the same confirm + match pipeline.' },
+        { title: 'Fix popup: MusicBrainz added to the auto-search cascade', desc: 'the Fix Track Match modal used to query only Spotify → Deezer → iTunes for the auto-search, leaving MusicBrainz out of the loop entirely — even for users with MusicBrainz set as their primary metadata source. now MB is part of the cascade. when MB is your primary, it gets queried first; otherwise it sits as the last fallback. catches niche / non-mainstream / canonical-with-diacritics recordings that the commercial sources miss. Discogs is intentionally absent — Discogs has no track-level search API.' },
     ],
     '2.5.5': [
         { date: 'May 17, 2026 — 2.5.5 release' },
diff --git a/webui/static/wishlist-tools.js b/webui/static/wishlist-tools.js
index ad705785..1a50bf19 100644
--- a/webui/static/wishlist-tools.js
+++ b/webui/static/wishlist-tools.js
@@ -214,12 +214,20 @@ async function searchDiscoveryFix() {
     }
     params.set('limit', '50');
 
-    // Use the user's active metadata source first, then fall back to others
+    // Use the user's active metadata source first, then fall back to others.
+    // MusicBrainz is included so users on MB-as-primary get MB queried first,
+    // and so MB is available as a fallback for fuzzy / niche / non-mainstream
+    // recordings that Spotify / Deezer / iTunes miss (different catalogues,
+    // different cover coverage). MB sits last by default because it's
+    // rate-limited to 1 req/sec — when it's the active primary the activeIdx
+    // reorder below moves it to the front. Discogs is intentionally absent —
+    // Discogs has no track-level search API (releases only).
     const activeSource = (currentMusicSourceName || 'Spotify').toLowerCase();
     const allSources = [
         { key: 'spotify', endpoint: '/api/spotify/search_tracks', label: 'Spotify' },
         { key: 'deezer', endpoint: '/api/deezer/search_tracks', label: 'Deezer' },
         { key: 'itunes', endpoint: '/api/itunes/search_tracks', label: 'iTunes' },
+        { key: 'musicbrainz', endpoint: '/api/musicbrainz/search_tracks', label: 'MusicBrainz' },
     ];
     // Put the active source first, keep others as fallbacks
     const activeIdx = allSources.findIndex(s => activeSource.includes(s.key));