diff --git a/phoneuse/android_mcp.py b/phoneuse/android_mcp.py
index 4e91dae..366f166 100644
--- a/phoneuse/android_mcp.py
+++ b/phoneuse/android_mcp.py
@@ -928,7 +928,8 @@ def _build_server():
 
         Full pipeline (runs on-device):
           text → edge-tts (MP3) → ffmpeg (16kHz PCM) → silk-wasm (SILK V3)
-          → SEND_VOICE_INTERNAL broadcast → WeChat DB insert (status=4, delivered)
+          → SEND_VOICE_CGI (internal uploadvoice cgi): audio uploaded + delivered to
+            recipient + sent bubble on the sender side. Verified end-to-end.
 
         **DO NOT implement SILK encoding yourself.** This tool handles the entire
         pipeline. Just pass the text you want spoken.
diff --git a/phoneuse/wechat_hook_send.py b/phoneuse/wechat_hook_send.py
index 21e2f0a..61250bf 100644
--- a/phoneuse/wechat_hook_send.py
+++ b/phoneuse/wechat_hook_send.py
@@ -472,7 +472,8 @@ def send_voice_tts_via_hook(
     """Generate TTS from *text* and send as WeChat voice message to *wxid*.
 
     End-to-end: edge-tts → ffmpeg (16kHz PCM) → silk-wasm (SILK V3) →
-    wechathook SEND_VOICE_INTERNAL broadcast → WeChat DB insert, status=4.
+    wechathook SEND_VOICE_CGI (internal uploadvoice cgi): uploads the audio + delivers
+    to the recipient AND inserts a sent bubble on the sender side. Verified end-to-end.
 
     Args:
         wxid:  talker id (e.g. 'filehelper'). IMPORTANT: test only with 'filehelper'.
@@ -498,10 +499,11 @@ def send_voice_tts_via_hook(
         except Exception as e:
             return {"status": "error", "error": f"TTS/SILK generation failed: {e}"}
 
-        # Delegate staging + broadcast to send_voice_via_hook (it handles both
-        # on-device and off-device paths). Must call while tmp dir is still alive.
-        result = send_voice_via_hook(wxid=wxid, silk_path=silk_path,
-                                     duration_ms=duration_ms)
+        # Delegate staging + broadcast to the fully-synthetic cgi sender (uploads audio
+        # inline via the internal uploadvoice cgi + delivers to the recipient + inserts a
+        # sent bubble). Must call while tmp dir is still alive.
+        result = send_voice_cgi_via_hook(wxid=wxid, silk_path=silk_path,
+                                         duration_ms=duration_ms)
     result["tts_text"] = text[:100]
     result["tts_voice"] = voice
     return result