add unit tests

Signed-off-by: Jaideep Rao <jrao@redhat.com>

client-sdks/stainless/openapi.yml

@@ -12002,18 +12002,9 @@ components:
           - web_search_preview
           - web_search_preview_2025_03_11
           - web_search_2025_08_26
-        search_context_size:
-          anyOf:
-          - type: string
-            pattern: ^low|medium|high$
-          - type: 'null'
-          default: medium
       type: object
       title: OpenAIResponseInputToolChoiceWebSearch
-      description: |-
+      description: Indicates that the model should use web search to generate a response
-        Indicates that the model should use web search to generate a response.
-
-        This is an alias for OpenAIResponseInputToolWebSearch used in tool_choice context.
     OpenAIResponseMessage-Input:
       properties:
         content:
@@ -12420,6 +12411,7 @@ components:
       title: AllowedToolsConfig
       type: object
     CustomToolConfig:
+      description: Custom tool configuration for OpenAI-compatible chat completion requests.
       properties:
         name:
           title: Name

docs/static/deprecated-llama-stack-spec.yaml

@@ -8996,18 +8996,9 @@ components:
           - web_search_preview
           - web_search_preview_2025_03_11
           - web_search_2025_08_26
-        search_context_size:
-          anyOf:
-          - type: string
-            pattern: ^low|medium|high$
-          - type: 'null'
-          default: medium
       type: object
       title: OpenAIResponseInputToolChoiceWebSearch
-      description: |-
+      description: Indicates that the model should use web search to generate a response
-        Indicates that the model should use web search to generate a response.
-
-        This is an alias for OpenAIResponseInputToolWebSearch used in tool_choice context.
     OpenAIResponseMessage-Input:
       properties:
         content:
@@ -9414,6 +9405,7 @@ components:
       title: AllowedToolsConfig
       type: object
     CustomToolConfig:
+      description: Custom tool configuration for OpenAI-compatible chat completion requests.
       properties:
         name:
           title: Name

docs/static/experimental-llama-stack-spec.yaml

@@ -7799,18 +7799,9 @@ components:
           - web_search_preview
           - web_search_preview_2025_03_11
           - web_search_2025_08_26
-        search_context_size:
-          anyOf:
-          - type: string
-            pattern: ^low|medium|high$
-          - type: 'null'
-          default: medium
       type: object
       title: OpenAIResponseInputToolChoiceWebSearch
-      description: |-
+      description: Indicates that the model should use web search to generate a response
-        Indicates that the model should use web search to generate a response.
-
-        This is an alias for OpenAIResponseInputToolWebSearch used in tool_choice context.
     OpenAIResponseMessage-Output:
       properties:
         content:
@@ -8148,6 +8139,7 @@ components:
       title: AllowedToolsConfig
       type: object
     CustomToolConfig:
+      description: Custom tool configuration for OpenAI-compatible chat completion requests.
       properties:
         name:
           title: Name

docs/static/llama-stack-spec.yaml

@@ -10438,18 +10438,9 @@ components:
           - web_search_preview
           - web_search_preview_2025_03_11
           - web_search_2025_08_26
-        search_context_size:
-          anyOf:
-          - type: string
-            pattern: ^low|medium|high$
-          - type: 'null'
-          default: medium
       type: object
       title: OpenAIResponseInputToolChoiceWebSearch
-      description: |-
+      description: Indicates that the model should use web search to generate a response
-        Indicates that the model should use web search to generate a response.
-
-        This is an alias for OpenAIResponseInputToolWebSearch used in tool_choice context.
     OpenAIResponseMessage-Input:
       properties:
         content:
@@ -10856,6 +10847,7 @@ components:
       title: AllowedToolsConfig
       type: object
     CustomToolConfig:
+      description: Custom tool configuration for OpenAI-compatible chat completion requests.
       properties:
         name:
           title: Name

docs/static/stainless-llama-stack-spec.yaml

@@ -12002,18 +12002,9 @@ components:
           - web_search_preview
           - web_search_preview_2025_03_11
           - web_search_2025_08_26
-        search_context_size:
-          anyOf:
-          - type: string
-            pattern: ^low|medium|high$
-          - type: 'null'
-          default: medium
       type: object
       title: OpenAIResponseInputToolChoiceWebSearch
-      description: |-
+      description: Indicates that the model should use web search to generate a response
-        Indicates that the model should use web search to generate a response.
-
-        This is an alias for OpenAIResponseInputToolWebSearch used in tool_choice context.
     OpenAIResponseMessage-Input:
       properties:
         content:
@@ -12420,6 +12411,7 @@ components:
       title: AllowedToolsConfig
       type: object
     CustomToolConfig:
+      description: Custom tool configuration for OpenAI-compatible chat completion requests.
       properties:
         name:
           title: Name

src/llama_stack/providers/inline/agents/meta_reference/responses/openai_responses.py

@@ -473,7 +473,7 @@ class OpenAIResponsesImpl:
             model=model,
             messages=messages,
             response_tools=tools,
-            responses_tool_choice=tool_choice,
+            tool_choice=tool_choice,
             temperature=temperature,
             response_format=response_format,
             tool_context=tool_context,

src/llama_stack/providers/inline/agents/meta_reference/responses/streaming.py

@@ -8,8 +8,8 @@ import uuid
 from collections.abc import AsyncIterator
 from typing import Any
 
-from opentelemetry import trace
 from openai.types.chat import ChatCompletionToolParam
+from opentelemetry import trace
 
 from llama_stack.log import get_logger
 from llama_stack.providers.utils.inference.prompt_adapter import interleaved_content_as_str
@@ -217,7 +217,7 @@ class StreamingResponseOrchestrator:
             output=self._clone_outputs(outputs),
             text=self.text,
             tools=self.ctx.available_tools(),
-            tool_choice=self.ctx.responses_tool_choice,
+            tool_choice=self.ctx.tool_choice,
             error=error,
             usage=self.accumulated_usage,
             instructions=self.instructions,
@@ -253,17 +253,18 @@ class StreamingResponseOrchestrator:
         async for stream_event in self._process_tools(output_messages):
             yield stream_event
 
-        if self.ctx.responses_tool_choice and len(self.ctx.chat_tools) > 0:
+        chat_tool_choice = None
-            chat_tool_choice = await _process_tool_choice(
+        if self.ctx.tool_choice and len(self.ctx.chat_tools) > 0:
+            processed_tool_choice = await _process_tool_choice(
                 self.ctx.chat_tools,
-                self.ctx.responses_tool_choice,
+                self.ctx.tool_choice,
                 self.server_label_to_tools,
             )
             # chat_tool_choice can be str, dict-like object, or None
-            if isinstance(chat_tool_choice, str):
+            if isinstance(processed_tool_choice, str | type(None)):
-                self.ctx.chat_tool_choice = chat_tool_choice
+                chat_tool_choice = processed_tool_choice
             else:
-                self.ctx.chat_tool_choice = chat_tool_choice.model_dump()
+                chat_tool_choice = processed_tool_choice.model_dump()
 
         n_iter = 0
         messages = self.ctx.messages.copy()
@@ -284,7 +285,7 @@ class StreamingResponseOrchestrator:
                     messages=messages,
                     # Pydantic models are dict-compatible but mypy treats them as distinct types
                     tools=self.ctx.chat_tools,  # type: ignore[arg-type]
-                    tool_choice=self.ctx.chat_tool_choice,
+                    tool_choice=chat_tool_choice,
                     stream=True,
                     temperature=self.ctx.temperature,
                     response_format=response_format,
@@ -363,8 +364,8 @@ class StreamingResponseOrchestrator:
                 n_iter += 1
                 # After first iteration, reset tool_choice to "auto" to let model decide freely
                 # based on tool results (prevents infinite loops when forcing specific tools)
-                if n_iter == 1 and self.ctx.chat_tool_choice:
+                if n_iter == 1 and chat_tool_choice:
-                    self.ctx.chat_tool_choice = "auto"
+                    chat_tool_choice = "auto"
                 if n_iter >= self.max_infer_iters:
                     logger.info(
                         f"Exiting inference loop since iteration count({n_iter}) exceeds {self.max_infer_iters=}"
@@ -1332,13 +1333,13 @@ class StreamingResponseOrchestrator:
 
 async def _process_tool_choice(
     chat_tools: list[ChatCompletionToolParam],
-    responses_tool_choice: OpenAIResponseInputToolChoice,
+    tool_choice: OpenAIResponseInputToolChoice,
     server_label_to_tools: dict[str, list[str]],
 ) -> str | OpenAIChatCompletionToolChoice | None:
     """Process and validate the OpenAI Responses tool choice and return the appropriate chat completion tool choice object.
 
     :param chat_tools: The list of chat tools to enforce tool choice against.
-    :param responses_tool_choice: The OpenAI Responses tool choice to process.
+    :param tool_choice: The OpenAI Responses tool choice to process.
     :param server_label_to_tools: A dictionary mapping server labels to the list of tools available on that server.
     :return: The appropriate chat completion tool choice object.
     """
@@ -1347,8 +1348,8 @@ async def _process_tool_choice(
     # Note: chat_tools contains dicts, not objects
     chat_tool_names = [tool["function"]["name"] for tool in chat_tools if tool["type"] == "function"]
 
-    if isinstance(responses_tool_choice, OpenAIResponseInputToolChoiceMode):
+    if isinstance(tool_choice, OpenAIResponseInputToolChoiceMode):
-        if responses_tool_choice.value == "required":
+        if tool_choice.value == "required":
             if len(chat_tool_names) == 0:
                 return None
 
@@ -1358,18 +1359,17 @@ async def _process_tool_choice(
                 mode="required",
             )
         # return other modes as is
-        return responses_tool_choice.value
+        return tool_choice.value
 
-    elif isinstance(responses_tool_choice, OpenAIResponseInputToolChoiceAllowedTools):
+    elif isinstance(tool_choice, OpenAIResponseInputToolChoiceAllowedTools):
         # ensure that specified tool choices are available in the chat tools, if not, remove them from the list
         final_tools = []
-        for tool in responses_tool_choice.tools:
+        for tool in tool_choice.tools:
-            tool_name = tool.get("name")
             match tool.get("type"):
                 case "function":
-                    final_tools.append({"type": "function", "function": {"name": tool_name}})
+                    final_tools.append({"type": "function", "function": {"name": tool.get("name")}})
                 case "custom":
-                    final_tools.append({"type": "custom", "custom": {"name": tool_name}})
+                    final_tools.append({"type": "custom", "custom": {"name": tool.get("name")}})
                 case "mcp":
                     mcp_tools = convert_mcp_tool_choice(
                         chat_tool_names, tool.get("server_label"), server_label_to_tools, None
@@ -1390,14 +1390,14 @@ async def _process_tool_choice(
 
         return OpenAIChatCompletionToolChoiceAllowedTools(
             tools=final_tools,
-            mode=responses_tool_choice.mode,
+            mode=tool_choice.mode,
         )
 
     else:
         # Handle specific tool choice by type
         # Each case validates the tool exists in chat_tools before returning
-        tool_name = responses_tool_choice.name if responses_tool_choice.name else None
+        tool_name = getattr(tool_choice, "name", None)
-        match responses_tool_choice:
+        match tool_choice:
             case OpenAIResponseInputToolChoiceCustomTool():
                 if tool_name and tool_name not in chat_tool_names:
                     logger.warning(f"Tool {tool_name} not found in chat tools")
@@ -1425,7 +1425,7 @@ async def _process_tool_choice(
             case OpenAIResponseInputToolChoiceMCPTool():
                 tool_choice = convert_mcp_tool_choice(
                     chat_tool_names,
-                    responses_tool_choice.server_label,
+                    tool_choice.server_label,
                     server_label_to_tools,
                     tool_name,
                 )

src/llama_stack/providers/inline/agents/meta_reference/responses/types.py

@@ -5,7 +5,7 @@
 # the root directory of this source tree.
 
 from dataclasses import dataclass
-from typing import Any, cast
+from typing import cast
 
 from openai.types.chat import ChatCompletionToolParam
 from pydantic import BaseModel
@@ -161,8 +161,7 @@ class ChatCompletionContext(BaseModel):
     temperature: float | None
     response_format: OpenAIResponseFormatParam
     tool_context: ToolContext | None
-    responses_tool_choice: OpenAIResponseInputToolChoice | None = None
+    tool_choice: OpenAIResponseInputToolChoice | None = None
-    chat_tool_choice: str | dict[str, Any] | None = None
     approval_requests: list[OpenAIResponseMCPApprovalRequest] = []
     approval_responses: dict[str, OpenAIResponseMCPApprovalResponse] = {}
 
@@ -175,7 +174,7 @@ class ChatCompletionContext(BaseModel):
         response_format: OpenAIResponseFormatParam,
         tool_context: ToolContext,
         inputs: list[OpenAIResponseInput] | str,
-        responses_tool_choice: OpenAIResponseInputToolChoice | None = None,
+        tool_choice: OpenAIResponseInputToolChoice | None = None,
     ):
         super().__init__(
             model=model,
@@ -184,7 +183,7 @@ class ChatCompletionContext(BaseModel):
             temperature=temperature,
             response_format=response_format,
             tool_context=tool_context,
-            responses_tool_choice=responses_tool_choice,
+            tool_choice=tool_choice,
         )
         if not isinstance(inputs, str):
             self.approval_requests = [input for input in inputs if input.type == "mcp_approval_request"]

src/llama_stack_api/inference.py

@@ -577,6 +577,11 @@ class OpenAIChatCompletionToolChoiceFunctionTool(BaseModel):
 
 @json_schema_type
 class CustomToolConfig(BaseModel):
+    """Custom tool configuration for OpenAI-compatible chat completion requests.
+
+    :param name: Name of the custom tool
+    """
+
     name: str
 
 

src/llama_stack_api/openai_responses.py

@@ -565,12 +565,19 @@ class OpenAIResponseInputToolChoiceFileSearch(BaseModel):
 
 
 @json_schema_type
-class OpenAIResponseInputToolChoiceWebSearch(OpenAIResponseInputToolWebSearch):
+class OpenAIResponseInputToolChoiceWebSearch(BaseModel):
-    """Indicates that the model should use web search to generate a response.
+    """Indicates that the model should use web search to generate a response
 
-    This is an alias for OpenAIResponseInputToolWebSearch used in tool_choice context.
+    :param type: Web search tool type variant to use
     """
 
+    type: (
+        Literal["web_search"]
+        | Literal["web_search_preview"]
+        | Literal["web_search_preview_2025_03_11"]
+        | Literal["web_search_2025_08_26"]
+    ) = "web_search"
+
 
 @json_schema_type
 class OpenAIResponseInputToolChoiceFunctionTool(BaseModel):

tests/unit/providers/agents/meta_reference/test_response_tool_context.py

@@ -5,9 +5,22 @@
 # the root directory of this source tree.
 
 
+from llama_stack.providers.inline.agents.meta_reference.responses.streaming import (
+    _process_tool_choice,
+)
 from llama_stack.providers.inline.agents.meta_reference.responses.types import ToolContext
-from llama_stack_api.openai_responses import (
+from llama_stack_api import (
     MCPListToolsTool,
+    OpenAIChatCompletionToolChoiceAllowedTools,
+    OpenAIChatCompletionToolChoiceCustomTool,
+    OpenAIChatCompletionToolChoiceFunctionTool,
+    OpenAIResponseInputToolChoiceAllowedTools,
+    OpenAIResponseInputToolChoiceCustomTool,
+    OpenAIResponseInputToolChoiceFileSearch,
+    OpenAIResponseInputToolChoiceFunctionTool,
+    OpenAIResponseInputToolChoiceMCPTool,
+    OpenAIResponseInputToolChoiceMode,
+    OpenAIResponseInputToolChoiceWebSearch,
     OpenAIResponseInputToolFileSearch,
     OpenAIResponseInputToolFunction,
     OpenAIResponseInputToolMCP,
@@ -181,3 +194,326 @@ class TestToolContext:
         assert len(context.previous_tool_listings) == 1
         assert len(context.previous_tool_listings[0].tools) == 1
         assert context.previous_tool_listings[0].server_label == "anotherlabel"
+
+
+class TestProcessToolChoice:
+    """Comprehensive test suite for _process_tool_choice function."""
+
+    def setup_method(self):
+        """Set up common test fixtures."""
+        self.chat_tools = [
+            {"type": "function", "function": {"name": "get_weather"}},
+            {"type": "function", "function": {"name": "calculate"}},
+            {"type": "function", "function": {"name": "file_search"}},
+            {"type": "function", "function": {"name": "web_search"}},
+        ]
+        self.server_label_to_tools = {
+            "mcp_server_1": ["mcp_tool_1", "mcp_tool_2"],
+            "mcp_server_2": ["mcp_tool_3"],
+        }
+
+    async def test_mode_auto(self):
+        """Test auto mode - should return 'auto' string."""
+        tool_choice = OpenAIResponseInputToolChoiceMode.auto
+        result = await _process_tool_choice(self.chat_tools, tool_choice, self.server_label_to_tools)
+        assert result == "auto"
+
+    async def test_mode_none(self):
+        """Test none mode - should return 'none' string."""
+        tool_choice = OpenAIResponseInputToolChoiceMode.none
+        result = await _process_tool_choice(self.chat_tools, tool_choice, self.server_label_to_tools)
+        assert result == "none"
+
+    async def test_mode_required_with_tools(self):
+        """Test required mode with available tools - should return AllowedTools with all function tools."""
+        tool_choice = OpenAIResponseInputToolChoiceMode.required
+        result = await _process_tool_choice(self.chat_tools, tool_choice, self.server_label_to_tools)
+
+        assert isinstance(result, OpenAIChatCompletionToolChoiceAllowedTools)
+        assert result.allowed_tools.mode == "required"
+        assert len(result.allowed_tools.tools) == 4
+        tool_names = [tool["function"]["name"] for tool in result.allowed_tools.tools]
+        assert "get_weather" in tool_names
+        assert "calculate" in tool_names
+        assert "file_search" in tool_names
+        assert "web_search" in tool_names
+
+    async def test_mode_required_without_tools(self):
+        """Test required mode without available tools - should return None."""
+        tool_choice = OpenAIResponseInputToolChoiceMode.required
+        result = await _process_tool_choice([], tool_choice, self.server_label_to_tools)
+        assert result is None
+
+    async def test_allowed_tools_function(self):
+        """Test allowed_tools with function tool types."""
+        tool_choice = OpenAIResponseInputToolChoiceAllowedTools(
+            mode="required",
+            tools=[
+                {"type": "function", "name": "get_weather"},
+                {"type": "function", "name": "calculate"},
+            ],
+        )
+        result = await _process_tool_choice(self.chat_tools, tool_choice, self.server_label_to_tools)
+
+        assert isinstance(result, OpenAIChatCompletionToolChoiceAllowedTools)
+        assert result.allowed_tools.mode == "required"
+        assert len(result.allowed_tools.tools) == 2
+        assert result.allowed_tools.tools[0]["function"]["name"] == "get_weather"
+        assert result.allowed_tools.tools[1]["function"]["name"] == "calculate"
+
+    async def test_allowed_tools_custom(self):
+        """Test allowed_tools with custom tool types."""
+        chat_tools = [{"type": "function", "function": {"name": "custom_tool_1"}}]
+        tool_choice = OpenAIResponseInputToolChoiceAllowedTools(
+            mode="auto",
+            tools=[{"type": "custom", "name": "custom_tool_1"}],
+        )
+        result = await _process_tool_choice(chat_tools, tool_choice, self.server_label_to_tools)
+
+        assert isinstance(result, OpenAIChatCompletionToolChoiceAllowedTools)
+        assert result.allowed_tools.mode == "auto"
+        assert len(result.allowed_tools.tools) == 1
+        assert result.allowed_tools.tools[0]["type"] == "custom"
+        assert result.allowed_tools.tools[0]["custom"]["name"] == "custom_tool_1"
+
+    async def test_allowed_tools_file_search(self):
+        """Test allowed_tools with file_search."""
+        tool_choice = OpenAIResponseInputToolChoiceAllowedTools(
+            mode="required",
+            tools=[{"type": "file_search"}],
+        )
+        result = await _process_tool_choice(self.chat_tools, tool_choice, self.server_label_to_tools)
+
+        assert isinstance(result, OpenAIChatCompletionToolChoiceAllowedTools)
+        assert len(result.allowed_tools.tools) == 1
+        assert result.allowed_tools.tools[0]["function"]["name"] == "file_search"
+
+    async def test_allowed_tools_web_search(self):
+        """Test allowed_tools with web_search."""
+        tool_choice = OpenAIResponseInputToolChoiceAllowedTools(
+            mode="required",
+            tools=[
+                {"type": "web_search_preview_2025_03_11"},
+                {"type": "web_search_2025_08_26"},
+                {"type": "web_search_preview"},
+            ],
+        )
+        result = await _process_tool_choice(self.chat_tools, tool_choice, self.server_label_to_tools)
+
+        assert isinstance(result, OpenAIChatCompletionToolChoiceAllowedTools)
+        assert len(result.allowed_tools.tools) == 3
+        assert result.allowed_tools.tools[0]["function"]["name"] == "web_search"
+        assert result.allowed_tools.tools[0]["type"] == "function"
+        assert result.allowed_tools.tools[1]["function"]["name"] == "web_search"
+        assert result.allowed_tools.tools[1]["type"] == "function"
+        assert result.allowed_tools.tools[2]["function"]["name"] == "web_search"
+        assert result.allowed_tools.tools[2]["type"] == "function"
+
+    async def test_allowed_tools_mcp_server_label(self):
+        """Test allowed_tools with MCP server label (no specific tool name)."""
+        chat_tools = [
+            {"type": "function", "function": {"name": "mcp_tool_1"}},
+            {"type": "function", "function": {"name": "mcp_tool_2"}},
+        ]
+        tool_choice = OpenAIResponseInputToolChoiceAllowedTools(
+            mode="required",
+            tools=[{"type": "mcp", "server_label": "mcp_server_1"}],
+        )
+        result = await _process_tool_choice(chat_tools, tool_choice, self.server_label_to_tools)
+
+        assert isinstance(result, OpenAIChatCompletionToolChoiceAllowedTools)
+        assert len(result.allowed_tools.tools) == 2
+        tool_names = [tool["function"]["name"] for tool in result.allowed_tools.tools]
+        assert "mcp_tool_1" in tool_names
+        assert "mcp_tool_2" in tool_names
+
+    async def test_allowed_tools_mixed_types(self):
+        """Test allowed_tools with mixed tool types."""
+        chat_tools = [
+            {"type": "function", "function": {"name": "get_weather"}},
+            {"type": "function", "function": {"name": "file_search"}},
+            {"type": "function", "function": {"name": "web_search"}},
+            {"type": "function", "function": {"name": "mcp_tool_1"}},
+        ]
+        tool_choice = OpenAIResponseInputToolChoiceAllowedTools(
+            mode="auto",
+            tools=[
+                {"type": "function", "name": "get_weather"},
+                {"type": "file_search"},
+                {"type": "web_search"},
+                {"type": "mcp", "server_label": "mcp_server_1"},
+            ],
+        )
+        result = await _process_tool_choice(chat_tools, tool_choice, self.server_label_to_tools)
+
+        assert isinstance(result, OpenAIChatCompletionToolChoiceAllowedTools)
+        # Should have: get_weather, file_search, web_search, mcp_tool_1, mcp_tool_2
+        assert len(result.allowed_tools.tools) >= 3
+
+    async def test_allowed_tools_invalid_type(self):
+        """Test allowed_tools with an unsupported tool type - should skip it."""
+        tool_choice = OpenAIResponseInputToolChoiceAllowedTools(
+            mode="required",
+            tools=[
+                {"type": "function", "name": "get_weather"},
+                {"type": "unsupported_type", "name": "bad_tool"},
+            ],
+        )
+        result = await _process_tool_choice(self.chat_tools, tool_choice, self.server_label_to_tools)
+
+        assert isinstance(result, OpenAIChatCompletionToolChoiceAllowedTools)
+        # Should only include the valid function tool
+        assert len(result.allowed_tools.tools) == 1
+        assert result.allowed_tools.tools[0]["function"]["name"] == "get_weather"
+
+    async def test_specific_custom_tool_valid(self):
+        """Test specific custom tool choice when tool exists."""
+        chat_tools = [{"type": "function", "function": {"name": "custom_tool"}}]
+        tool_choice = OpenAIResponseInputToolChoiceCustomTool(name="custom_tool")
+        result = await _process_tool_choice(chat_tools, tool_choice, self.server_label_to_tools)
+
+        assert isinstance(result, OpenAIChatCompletionToolChoiceCustomTool)
+        assert result.custom.name == "custom_tool"
+
+    async def test_specific_custom_tool_invalid(self):
+        """Test specific custom tool choice when tool doesn't exist - should return None."""
+        tool_choice = OpenAIResponseInputToolChoiceCustomTool(name="nonexistent_tool")
+        result = await _process_tool_choice(self.chat_tools, tool_choice, self.server_label_to_tools)
+        assert result is None
+
+    async def test_specific_function_tool_valid(self):
+        """Test specific function tool choice when tool exists."""
+        tool_choice = OpenAIResponseInputToolChoiceFunctionTool(name="get_weather")
+        result = await _process_tool_choice(self.chat_tools, tool_choice, self.server_label_to_tools)
+
+        assert isinstance(result, OpenAIChatCompletionToolChoiceFunctionTool)
+        assert result.function.name == "get_weather"
+
+    async def test_specific_function_tool_invalid(self):
+        """Test specific function tool choice when tool doesn't exist - should return None."""
+        tool_choice = OpenAIResponseInputToolChoiceFunctionTool(name="nonexistent_function")
+        result = await _process_tool_choice(self.chat_tools, tool_choice, self.server_label_to_tools)
+        assert result is None
+
+    async def test_specific_file_search_valid(self):
+        """Test file_search tool choice when available."""
+        tool_choice = OpenAIResponseInputToolChoiceFileSearch()
+        result = await _process_tool_choice(self.chat_tools, tool_choice, self.server_label_to_tools)
+
+        assert isinstance(result, OpenAIChatCompletionToolChoiceFunctionTool)
+        assert result.function.name == "file_search"
+
+    async def test_specific_file_search_invalid(self):
+        """Test file_search tool choice when not available - should return None."""
+        chat_tools = [{"type": "function", "function": {"name": "get_weather"}}]
+        tool_choice = OpenAIResponseInputToolChoiceFileSearch()
+        result = await _process_tool_choice(chat_tools, tool_choice, self.server_label_to_tools)
+        assert result is None
+
+    async def test_specific_web_search_valid(self):
+        """Test web_search tool choice when available."""
+        tool_choice = OpenAIResponseInputToolChoiceWebSearch()
+        result = await _process_tool_choice(self.chat_tools, tool_choice, self.server_label_to_tools)
+
+        assert isinstance(result, OpenAIChatCompletionToolChoiceFunctionTool)
+        assert result.function.name == "web_search"
+
+    async def test_specific_web_search_invalid(self):
+        """Test web_search tool choice when not available - should return None."""
+        chat_tools = [{"type": "function", "function": {"name": "get_weather"}}]
+        tool_choice = OpenAIResponseInputToolChoiceWebSearch()
+        result = await _process_tool_choice(chat_tools, tool_choice, self.server_label_to_tools)
+        assert result is None
+
+    async def test_specific_mcp_tool_with_name(self):
+        """Test MCP tool choice with specific tool name."""
+        chat_tools = [{"type": "function", "function": {"name": "mcp_tool_1"}}]
+        tool_choice = OpenAIResponseInputToolChoiceMCPTool(
+            server_label="mcp_server_1",
+            name="mcp_tool_1",
+        )
+        result = await _process_tool_choice(chat_tools, tool_choice, self.server_label_to_tools)
+
+        assert isinstance(result, OpenAIChatCompletionToolChoiceFunctionTool)
+        assert result.function.name == "mcp_tool_1"
+
+    async def test_specific_mcp_tool_with_name_not_in_chat_tools(self):
+        """Test MCP tool choice with specific tool name that doesn't exist in chat_tools."""
+        chat_tools = [{"type": "function", "function": {"name": "other_tool"}}]
+        tool_choice = OpenAIResponseInputToolChoiceMCPTool(
+            server_label="mcp_server_1",
+            name="mcp_tool_1",
+        )
+        result = await _process_tool_choice(chat_tools, tool_choice, self.server_label_to_tools)
+        assert result is None
+
+    async def test_specific_mcp_tool_server_label_only(self):
+        """Test MCP tool choice with only server label (no specific tool name)."""
+        chat_tools = [
+            {"type": "function", "function": {"name": "mcp_tool_1"}},
+            {"type": "function", "function": {"name": "mcp_tool_2"}},
+        ]
+        tool_choice = OpenAIResponseInputToolChoiceMCPTool(server_label="mcp_server_1")
+        result = await _process_tool_choice(chat_tools, tool_choice, self.server_label_to_tools)
+
+        assert isinstance(result, OpenAIChatCompletionToolChoiceAllowedTools)
+        assert result.allowed_tools.mode == "required"
+        assert len(result.allowed_tools.tools) == 2
+        tool_names = [tool["function"]["name"] for tool in result.allowed_tools.tools]
+        assert "mcp_tool_1" in tool_names
+        assert "mcp_tool_2" in tool_names
+
+    async def test_specific_mcp_tool_unknown_server(self):
+        """Test MCP tool choice with unknown server label."""
+        tool_choice = OpenAIResponseInputToolChoiceMCPTool(
+            server_label="unknown_server",
+            name="some_tool",
+        )
+        result = await _process_tool_choice(self.chat_tools, tool_choice, self.server_label_to_tools)
+        # Should return None because server not found
+        assert result is None
+
+    async def test_empty_chat_tools(self):
+        """Test with empty chat_tools list."""
+        tool_choice = OpenAIResponseInputToolChoiceFunctionTool(name="get_weather")
+        result = await _process_tool_choice([], tool_choice, self.server_label_to_tools)
+        assert result is None
+
+    async def test_empty_server_label_to_tools(self):
+        """Test with empty server_label_to_tools mapping."""
+        tool_choice = OpenAIResponseInputToolChoiceMCPTool(server_label="mcp_server_1")
+        result = await _process_tool_choice(self.chat_tools, tool_choice, {})
+        # Should handle gracefully
+        assert result is None or isinstance(result, OpenAIChatCompletionToolChoiceAllowedTools)
+
+    async def test_allowed_tools_empty_list(self):
+        """Test allowed_tools with empty tools list."""
+        tool_choice = OpenAIResponseInputToolChoiceAllowedTools(mode="auto", tools=[])
+        result = await _process_tool_choice(self.chat_tools, tool_choice, self.server_label_to_tools)
+
+        assert isinstance(result, OpenAIChatCompletionToolChoiceAllowedTools)
+        assert len(result.allowed_tools.tools) == 0
+
+    async def test_mcp_tool_multiple_servers(self):
+        """Test MCP tool choice with multiple server labels."""
+        chat_tools = [
+            {"type": "function", "function": {"name": "mcp_tool_1"}},
+            {"type": "function", "function": {"name": "mcp_tool_2"}},
+            {"type": "function", "function": {"name": "mcp_tool_3"}},
+        ]
+        server_label_to_tools = {
+            "server_a": ["mcp_tool_1"],
+            "server_b": ["mcp_tool_2", "mcp_tool_3"],
+        }
+
+        # Test server_a
+        tool_choice_a = OpenAIResponseInputToolChoiceMCPTool(server_label="server_a")
+        result_a = await _process_tool_choice(chat_tools, tool_choice_a, server_label_to_tools)
+        assert isinstance(result_a, OpenAIChatCompletionToolChoiceAllowedTools)
+        assert len(result_a.allowed_tools.tools) == 1
+
+        # Test server_b
+        tool_choice_b = OpenAIResponseInputToolChoiceMCPTool(server_label="server_b")
+        result_b = await _process_tool_choice(chat_tools, tool_choice_b, server_label_to_tools)
+        assert isinstance(result_b, OpenAIChatCompletionToolChoiceAllowedTools)
+        assert len(result_b.allowed_tools.tools) == 2