diff --git a/src/google/adk/plugins/bigquery_agent_analytics_plugin.py b/src/google/adk/plugins/bigquery_agent_analytics_plugin.py
index 0db62f1be6..d031a5f13d 100644
--- a/src/google/adk/plugins/bigquery_agent_analytics_plugin.py
+++ b/src/google/adk/plugins/bigquery_agent_analytics_plugin.py
@@ -8191,13 +8191,29 @@ async def after_tool_callback(
     span_id, duration = TraceManager.pop_span()
     parent_span_id, _ = TraceManager.get_current_span_and_parent()
 
+    # MCP CallToolResult (isError) and plugin-rewritten failures
+    # (error_details) reach here as ordinary results. Classify them as
+    # TOOL_ERROR so status/error_message match on_tool_error_callback;
+    # otherwise audits treat them as successes.
+    error_message = None
+    if isinstance(result, dict):
+      if result.get("error_details"):
+        error_message = str(result["error_details"])
+      elif result.get("isError") or result.get("is_error"):
+        content = result.get("content")
+        first = content[0] if isinstance(content, list) and content else None
+        text = first.get("text") if isinstance(first, dict) else None
+        error_message = str(text) if text else "Tool returned an error"
+
     event_data = EventData(
         latency_ms=duration,
         span_id_override=span_id,
         parent_span_id_override=parent_span_id,
+        status="ERROR" if error_message is not None else "OK",
+        error_message=error_message,
     )
     await self._log_event(
-        "TOOL_COMPLETED",
+        "TOOL_ERROR" if error_message is not None else "TOOL_COMPLETED",
         tool_context,
         raw_content=content_dict,
         is_truncated=is_truncated,
@@ -8211,7 +8227,10 @@ async def after_tool_callback(
     # args (the final-answer payload the model supplied) as AGENT_RESPONSE so
     # the visible response text is captured.  Opt-in via
     # ``config.final_response_tool_names`` (empty by default).
-    if tool.name in self.config.final_response_tool_names:
+    if (
+        error_message is None
+        and tool.name in self.config.final_response_tool_names
+    ):
       args_truncated, args_is_truncated = _recursive_smart_truncate(
           tool_args, self.config.max_content_length
       )
diff --git a/tests/unittests/plugins/test_bigquery_agent_analytics_plugin.py b/tests/unittests/plugins/test_bigquery_agent_analytics_plugin.py
index 854d47b031..1257b2e9e1 100644
--- a/tests/unittests/plugins/test_bigquery_agent_analytics_plugin.py
+++ b/tests/unittests/plugins/test_bigquery_agent_analytics_plugin.py
@@ -2119,6 +2119,52 @@ async def test_after_tool_callback_logs_correctly(
     assert content_dict["tool"] == "MyTool"
     assert content_dict["result"] == {"res": "success"}
 
+  @pytest.mark.asyncio
+  @pytest.mark.parametrize(
+      ("result", "expected_error_message"),
+      [
+          (
+              {
+                  "content": [
+                      {"type": "text", "text": "Git reset to remote failed."}
+                  ],
+                  "isError": True,
+              },
+              "Git reset to remote failed.",
+          ),
+          (
+              {"error_details": "MCP request failed with code 403"},
+              "MCP request failed with code 403",
+          ),
+      ],
+  )
+  async def test_after_tool_callback_logs_error_bearing_result_as_tool_error(
+      self, bq_plugin_inst, tool_context, result, expected_error_message
+  ):
+    """Error-bearing tool results are recorded as TOOL_ERROR, not TOOL_COMPLETED."""
+    mock_tool = mock.create_autospec(
+        base_tool_lib.BaseTool, instance=True, spec_set=True
+    )
+    type(mock_tool).name = mock.PropertyMock(return_value="MyTool")
+    type(mock_tool).description = mock.PropertyMock(return_value="Description")
+    log_event = mock.AsyncMock()
+    bq_plugin_inst._log_event = log_event
+    bigquery_agent_analytics_plugin.TraceManager.push_span(tool_context)
+
+    await bq_plugin_inst.after_tool_callback(
+        tool=mock_tool,
+        tool_args={"arg1": "val1"},
+        tool_context=tool_context,
+        result=result,
+    )
+
+    log_event.assert_awaited_once()
+    assert log_event.await_args.args[0] == "TOOL_ERROR"
+    event_data = log_event.await_args.kwargs["event_data"]
+    assert event_data.status == "ERROR"
+    assert event_data.error_message == expected_error_message
+    assert log_event.await_args.kwargs["raw_content"]["tool"] == "MyTool"
+
   @pytest.mark.asyncio
   async def test_after_tool_callback_no_state_delta_logging(
       self, bq_plugin_inst, mock_write_client, tool_context, dummy_arrow_schema