diff --git a/litellm/llms/anthropic/pass_through/adapters/transformation.py b/litellm/llms/anthropic/pass_through/adapters/transformation.py
index 3aad0be1913..e42cadbe3af 100644
--- a/litellm/llms/anthropic/pass_through/adapters/transformation.py
+++ b/litellm/llms/anthropic/pass_through/adapters/transformation.py
@@ -529,10 +529,16 @@ class LiteLLMAnthropicMessagesAdapter:
                                 self._add_cache_control_if_applicable(content, doc_obj, model)
                                 new_user_content_list.append(doc_obj)
                         elif content.get("type") == "tool_result":
+                            raw_tool_result_content: Final = self._tool_result_content(content.get("content"))
+                            tool_result_content: Final = (
+                                self._mark_tool_result_as_error(raw_tool_result_content)
+                                if content.get("is_error")
+                                else raw_tool_result_content
+                            )
                             tool_result = ChatCompletionToolMessage(
                                 role="tool",
                                 tool_call_id=content.get("tool_use_id", ""),
-                                content=self._tool_result_content(content.get("content")),
+                                content=tool_result_content,
                             )
                             self._add_cache_control_if_applicable(content, tool_result, model)
                             tool_message_list.append(tool_result)
@@ -1321,6 +1327,20 @@ class LiteLLMAnthropicMessagesAdapter:
             case _:
                 return list(parts)
 
+    @staticmethod
+    def _mark_tool_result_as_error(content: ToolResultContent) -> ToolResultContent:
+        """Keep Anthropic's `is_error` signal visible.
+
+        Chat Completions tool messages have no error field, so encode the
+        flag into the content itself; models and downstream provider
+        mappings can then still tell the tool call failed.
+        """
+        if isinstance(content, str):
+            return "[tool error] " + content if content else "[tool error]"
+        if isinstance(content, list):
+            return [ChatCompletionTextObject(type="text", text="[tool error] "), *content]
+        return content
+
     def _tool_result_part(self, item: object) -> ToolMessageContentPart | None:
         if isinstance(item, str):
             return ChatCompletionTextObject(type="text", text=item)
diff --git a/tests/unit/llms/anthropic/pass_through/adapters/test_anthropic_experimental_pass_through_adapters_transformation.py b/tests/unit/llms/anthropic/pass_through/adapters/test_anthropic_experimental_pass_through_adapters_transformation.py
index 9e0bd5c60b8..70c77eefdf1 100644
--- a/tests/unit/llms/anthropic/pass_through/adapters/test_anthropic_experimental_pass_through_adapters_transformation.py
+++ b/tests/unit/llms/anthropic/pass_through/adapters/test_anthropic_experimental_pass_through_adapters_transformation.py
@@ -4712,6 +4712,106 @@ def test_tool_result_without_translatable_content_still_answers_its_tool_use(too
     assert result[0]["role"] == "assistant"
 
 
+def test_tool_result_is_error_is_encoded_in_string_content():
+    adapter = LiteLLMAnthropicMessagesAdapter()
+
+    result = adapter.translate_anthropic_messages_to_openai(
+        messages=[
+            _anthropic_tool_use_turn("toolu_01"),
+            {
+                "role": "user",
+                "content": [
+                    {
+                        "type": "tool_result",
+                        "tool_use_id": "toolu_01",
+                        "content": "deploy failed",
+                        "is_error": True,
+                    }
+                ],
+            },
+        ]
+    )
+
+    assert result[1] == {
+        "role": "tool",
+        "tool_call_id": "toolu_01",
+        "content": "[tool error] deploy failed",
+    }
+
+
+def test_tool_result_is_error_puts_the_marker_first_in_part_lists():
+    adapter = LiteLLMAnthropicMessagesAdapter()
+
+    result = adapter.translate_anthropic_messages_to_openai(
+        messages=[
+            _anthropic_tool_use_turn("toolu_01"),
+            {
+                "role": "user",
+                "content": [
+                    {
+                        "type": "tool_result",
+                        "tool_use_id": "toolu_01",
+                        "content": [{"type": "text", "text": "boom"}, _tool_reference_block()],
+                        "is_error": True,
+                    }
+                ],
+            },
+        ]
+    )
+
+    assert result[1]["content"] == [
+        {"type": "text", "text": "[tool error] "},
+        {"type": "text", "text": "boom"},
+        {"type": "tool_reference", "tool_name": "WebFetch"},
+    ]
+
+
+def test_tool_result_is_error_with_empty_content_keeps_the_marker():
+    adapter = LiteLLMAnthropicMessagesAdapter()
+
+    result = adapter.translate_anthropic_messages_to_openai(
+        messages=[
+            _anthropic_tool_use_turn("toolu_01"),
+            {
+                "role": "user",
+                "content": [
+                    {
+                        "type": "tool_result",
+                        "tool_use_id": "toolu_01",
+                        "content": "",
+                        "is_error": True,
+                    }
+                ],
+            },
+        ]
+    )
+
+    assert result[1]["content"] == "[tool error]"
+
+
+def test_tool_result_without_is_error_keeps_content_unchanged():
+    adapter = LiteLLMAnthropicMessagesAdapter()
+
+    result = adapter.translate_anthropic_messages_to_openai(
+        messages=[
+            _anthropic_tool_use_turn("toolu_01"),
+            {
+                "role": "user",
+                "content": [
+                    {
+                        "type": "tool_result",
+                        "tool_use_id": "toolu_01",
+                        "content": "deploy succeeded",
+                        "is_error": False,
+                    }
+                ],
+            },
+        ]
+    )
+
+    assert result[1]["content"] == "deploy succeeded"
+
+
 def _openai_response_with_usage(usage: Usage) -> ModelResponse:
     return ModelResponse(
         id="resp_web_search",