diff --git a/litellm/llms/anthropic/messages/responses_adapters/transformation.py b/litellm/llms/anthropic/messages/responses_adapters/transformation.py
index 110e9fc40ff..40e335dfe17 100644
--- a/litellm/llms/anthropic/messages/responses_adapters/transformation.py
+++ b/litellm/llms/anthropic/messages/responses_adapters/transformation.py
@@ -25,7 +25,7 @@ from litellm.litellm_core_utils.prompt_templates.reasoning_utils import (
 from litellm.litellm_core_utils.reasoning_effort_utils import (
     reasoning_effort_from_thinking,
 )
-from litellm.llms.anthropic.messages.convertible_blocks import tool_result_text_parts
+from litellm.llms.anthropic.messages.convertible_blocks import tool_result_text_part
 from litellm.llms.anthropic.messages.native.utils import (
     refusal_stop_details,
     responses_output_refusal_text,
@@ -60,6 +60,22 @@ REASONING_SUMMARY_PART_SEPARATOR: Final = "\n\n"
 RESPONSES_INCLUDE_ENCRYPTED_REASONING: Final = "reasoning.encrypted_content"
 
 
+def _responses_tool_result_part(block: object) -> str | None:
+    match block:
+        case {"type": "tool_reference", "tool_name": str(tool_name)}:
+            return f"[Loaded tool: {tool_name}]"
+        case {"type": "tool_reference"}:
+            return "[Loaded tool: ]"
+        case _:
+            return tool_result_text_part(block)
+
+
+def _responses_tool_result_text_parts(tool_result: Mapping[str, object]) -> tuple[str, ...]:
+    content: Final = tool_result.get("content")
+    parts: Final = map(_responses_tool_result_part, content if isinstance(content, list) else ())
+    return tuple(part for part in parts if part is not None)
+
+
 class LiteLLMAnthropicToResponsesAPIAdapter:
     """
     Converts Anthropic /v1/messages requests to OpenAI Responses API format and
@@ -387,7 +403,7 @@ class LiteLLMAnthropicToResponsesAPIAdapter:
                             elif isinstance(inner, str):
                                 output_text = inner
                             elif isinstance(inner, list):
-                                output_text = "\n".join(tool_result_text_parts(tool_result_block))
+                                output_text = "\n".join(_responses_tool_result_text_parts(tool_result_block))
                                 if covered_image_parts or later_image_parts:
                                     output_text = (
                                         f"{output_text}\n{TOOL_RESULT_IMAGE_PLACEHOLDER}"
diff --git a/tests/unit/llms/anthropic/messages/responses_adapters/test_responses_adapters_transformation.py b/tests/unit/llms/anthropic/messages/responses_adapters/test_responses_adapters_transformation.py
index dbb8c925fa0..079f974e439 100644
--- a/tests/unit/llms/anthropic/messages/responses_adapters/test_responses_adapters_transformation.py
+++ b/tests/unit/llms/anthropic/messages/responses_adapters/test_responses_adapters_transformation.py
@@ -490,6 +490,33 @@ class TestTranslateMessagesToResponsesInput:
         result = _translate_messages(messages)
         assert result[0]["output"] == "Line 1\nLine 2"
 
+    def test_user_tool_result_list_content_preserves_tool_references(self):
+        messages = [
+            {
+                "role": "user",
+                "content": [
+                    {
+                        "type": "tool_result",
+                        "tool_use_id": "call_search",
+                        "content": [
+                            {"type": "text", "text": "Available:"},
+                            {"type": "tool_reference", "tool_name": "get_weather"},
+                        ],
+                    }
+                ],
+            }
+        ]
+
+        result = _translate_messages(messages)
+
+        assert result == [
+            {
+                "type": "function_call_output",
+                "call_id": "call_search",
+                "output": "Available:\n[Loaded tool: get_weather]",
+            }
+        ]
+
     def test_user_tool_result_null_content(self):
         """tool_result with null content becomes empty string output."""
         messages = [
diff --git a/tests/unit/llms/anthropic/messages/test_convertible_blocks.py b/tests/unit/llms/anthropic/messages/test_convertible_blocks.py
index 5fb6cc32dfb..4bad6cfb754 100644
--- a/tests/unit/llms/anthropic/messages/test_convertible_blocks.py
+++ b/tests/unit/llms/anthropic/messages/test_convertible_blocks.py
@@ -136,7 +136,6 @@ _DROPPED_INSIDE_TOOL_RESULT: Final[tuple[tuple[Mapping[str, object], str, str],
         "text-notes",
         "document with a text source inside a tool_result",
     ),
-    ({"type": "tool_reference", "tool_name": "find_docs"}, "find_docs", "tool_reference inside a tool_result"),
 )
 
 
@@ -158,6 +157,22 @@ def test_tool_result_content_both_counters_drop_is_named(inner: Mapping[str, obj
     assert unconvertible_anthropic_blocks(_tool_turn(inner)) == (label,)
 
 
+def test_tool_reference_count_guard_rejects_when_only_responses_preserves_it() -> None:
+    inner: Final = {"type": "tool_reference", "tool_name": "find_docs"}
+    gemini, openai = _sent_by_each_counter(inner)
+
+    assert "find_docs" not in gemini
+    assert _JSON_OBJECTS.validate_json(openai) == [
+        {"type": "function_call", "call_id": "toolu_01", "name": "lookup", "arguments": "{}"},
+        {"type": "function_call_output", "call_id": "toolu_01", "output": "[Loaded tool: find_docs]"},
+    ]
+    assert unconvertible_anthropic_blocks(_tool_turn(inner)) == ("tool_reference inside a tool_result",)
+    assert unconvertible_blocks_error(_tool_turn(inner), "anthropic") == (
+        "count_tokens can't convert these content blocks for this model, so the count would leave them out: "
+        "tool_reference inside a tool_result"
+    )
+
+
 def _replayed_web_search(encrypted_content: str) -> tuple[Mapping[str, object], ...]:
     return (
         {"role": "user", "content": "What is new?"},