diff --git a/src/google/adk/models/lite_llm.py b/src/google/adk/models/lite_llm.py
index 14047398..975acc31 100644
--- a/src/google/adk/models/lite_llm.py
+++ b/src/google/adk/models/lite_llm.py
@@ -104,6 +104,11 @@ _SUPPORTED_FILE_CONTENT_MIME_TYPES = frozenset({
 # Providers that require file_id instead of inline file_data
 _FILE_ID_REQUIRED_PROVIDERS = frozenset({"openai", "azure"})
 
+_MISSING_TOOL_RESULT_MESSAGE = (
+    "Error: Missing tool result (tool execution may have been interrupted "
+    "before a response was recorded)."
+)
+
 
 def _get_provider_from_model(model: str) -> str:
   """Extracts the provider name from a LiteLLM model string.
@@ -516,6 +521,65 @@ async def _content_to_message_param(
     )
 
 
+def _ensure_tool_results(messages: List[Message]) -> List[Message]:
+  """Insert placeholder tool messages for missing tool results.
+
+  LiteLLM-backed providers like OpenAI and Anthropic reject histories where an
+  assistant tool call is not followed by tool responses before the next
+  non-tool message. This helps recover from interrupted tool execution.
+  """
+  if not messages:
+    return messages
+
+  healed_messages: List[Message] = []
+  pending_tool_call_ids: List[str] = []
+
+  for message in messages:
+    role = message.get("role")
+    if pending_tool_call_ids and role != "tool":
+      logger.warning(
+          "Missing tool results for tool_call_id(s): %s",
+          pending_tool_call_ids,
+      )
+      healed_messages.extend(
+          ChatCompletionToolMessage(
+              role="tool",
+              tool_call_id=tool_call_id,
+              content=_MISSING_TOOL_RESULT_MESSAGE,
+          )
+          for tool_call_id in pending_tool_call_ids
+      )
+      pending_tool_call_ids = []
+
+    if role == "assistant":
+      tool_calls = message.get("tool_calls") or []
+      pending_tool_call_ids = [
+          tool_call.get("id") for tool_call in tool_calls if tool_call.get("id")
+      ]
+    elif role == "tool":
+      tool_call_id = message.get("tool_call_id")
+      if tool_call_id in pending_tool_call_ids:
+        pending_tool_call_ids.remove(tool_call_id)
+
+    healed_messages.append(message)
+
+  if pending_tool_call_ids:
+    logger.warning(
+        "Missing tool results for tool_call_id(s): %s",
+        pending_tool_call_ids,
+    )
+    healed_messages.extend(
+        ChatCompletionToolMessage(
+            role="tool",
+            tool_call_id=tool_call_id,
+            content=_MISSING_TOOL_RESULT_MESSAGE,
+        )
+        for tool_call_id in pending_tool_call_ids
+    )
+
+  return healed_messages
+
+
 async def _get_content(
     parts: Iterable[types.Part],
     *,
@@ -1266,6 +1330,7 @@ async def _get_completion_inputs(
             content=llm_request.config.system_instruction,
         ),
     )
+  messages = _ensure_tool_results(messages)
 
   # 2. Convert tool declarations
   tools: Optional[List[Dict]] = None
diff --git a/tests/unittests/models/test_litellm.py b/tests/unittests/models/test_litellm.py
index 4cf0329a..ca36966c 100644
--- a/tests/unittests/models/test_litellm.py
+++ b/tests/unittests/models/test_litellm.py
@@ -32,6 +32,7 @@ from google.adk.models.lite_llm import _get_completion_inputs
 from google.adk.models.lite_llm import _get_content
 from google.adk.models.lite_llm import _get_provider_from_model
 from google.adk.models.lite_llm import _message_to_generate_content_response
+from google.adk.models.lite_llm import _MISSING_TOOL_RESULT_MESSAGE
 from google.adk.models.lite_llm import _model_response_to_chunk
 from google.adk.models.lite_llm import _model_response_to_generate_content_response
 from google.adk.models.lite_llm import _parse_tool_calls_from_text
@@ -470,6 +471,43 @@ async def test_get_completion_inputs_uses_passed_model_for_gemini_format():
   assert "response_schema" in response_format
 
 
+@pytest.mark.asyncio
+async def test_get_completion_inputs_inserts_missing_tool_results():
+  user_content = types.Content(
+      role="user", parts=[types.Part.from_text(text="Hi")]
+  )
+  assistant_content = types.Content(
+      role="assistant",
+      parts=[
+          types.Part.from_text(text="Calling tool."),
+          types.Part.from_function_call(
+              name="get_weather", args={"location": "Seoul"}
+          ),
+      ],
+  )
+  assistant_content.parts[1].function_call.id = "tool_call_1"
+  followup_user = types.Content(
+      role="user", parts=[types.Part.from_text(text="Next question.")]
+  )
+
+  llm_request = LlmRequest(
+      contents=[user_content, assistant_content, followup_user]
+  )
+  messages, _, _, _ = await _get_completion_inputs(
+      llm_request, model="openai/gpt-4o"
+  )
+
+  assert [message["role"] for message in messages] == [
+      "user",
+      "assistant",
+      "tool",
+      "user",
+  ]
+  tool_message = messages[2]
+  assert tool_message["tool_call_id"] == "tool_call_1"
+  assert tool_message["content"] == _MISSING_TOOL_RESULT_MESSAGE
+
+
 def test_schema_to_dict_filters_none_enum_values():
   # Use model_construct to bypass strict enum validation.
   top_level_schema = types.Schema.model_construct(