[Tool Parser][1/3] Pass tools to ToolParser constructor (#38029)

Signed-off-by: sfeng33 <4florafeng@gmail.com>

vllm/entrypoints/openai/chat_completion/serving.py

@@ -565,7 +565,7 @@ class OpenAIServingChat(OpenAIServing):
                     )
 
                 tool_parsers: list[ToolParser | None] = [
-                    self.tool_parser(tokenizer)
+                    self.tool_parser(tokenizer, request.tools)
                 ] * num_choices
             else:
                 tool_parsers = [None] * num_choices
@@ -1331,7 +1331,7 @@ class OpenAIServingChat(OpenAIServing):
                             "Tokenizer not available when `skip_tokenizer_init=True`"
                         )
 
-                    tool_parser = self.tool_parser(tokenizer)
+                    tool_parser = self.tool_parser(tokenizer, request.tools)
                     # NOTE: We use token_ids for openai tool parser
                     tool_call_info = tool_parser.extract_tool_calls(
                         "",

vllm/entrypoints/openai/engine/serving.py

@@ -925,7 +925,7 @@ class OpenAIServing:
 
             # Automatic Tool Call Parsing
             try:
-                tool_parser = tool_parser_cls(tokenizer)
+                tool_parser = tool_parser_cls(tokenizer, request.tools)
             except RuntimeError as e:
                 logger.exception("Error in tool parser creation.")
                 raise e

vllm/entrypoints/openai/parser/responses_parser.py

@@ -52,7 +52,7 @@ class ResponsesParser:
         self.reasoning_parser_instance = reasoning_parser_cls(tokenizer)
         self.tool_parser_instance = None
         if tool_parser_cls is not None:
-            self.tool_parser_instance = tool_parser_cls(tokenizer)
+            self.tool_parser_instance = tool_parser_cls(tokenizer, request.tools)
 
         # Store the last finish_reason to determine response status
         self.finish_reason: str | None = None

vllm/entrypoints/openai/responses/serving.py

@@ -1344,7 +1344,7 @@ class OpenAIServingResponses(OpenAIServing):
             reasoning_parser = self.parser.reasoning_parser_cls(tokenizer)
         tool_parser = None
         if self.parser and self.parser.tool_parser_cls:
-            tool_parser = self.parser.tool_parser_cls(tokenizer)
+            tool_parser = self.parser.tool_parser_cls(tokenizer, request.tools)
         reasoning_ended = False
         tool_call_text_started = False
         previous_text = ""

vllm/entrypoints/serve/render/serving.py

@@ -545,6 +545,8 @@ class OpenAIServingRender:
                     )
                     raise NotImplementedError(msg)
                 tokenizer = renderer.get_tokenizer()
-                request = tool_parser(tokenizer).adjust_request(request=request)  # type: ignore[arg-type]
+                request = tool_parser(tokenizer, request.tools).adjust_request(
+                    request=request  # type: ignore[arg-type]
+                )
 
         return conversation, [engine_input]

vllm/tool_parsers/abstract_tool_parser.py

@@ -5,13 +5,18 @@ import importlib
 import os
 from collections.abc import Callable, Sequence
 from functools import cached_property
+from typing import TypeAlias
 
 from openai.types.responses import (
     ResponseFormatTextJSONSchemaConfig,
     ResponseTextConfig,
 )
+from openai.types.responses.tool import Tool as ResponsesTool
 
-from vllm.entrypoints.openai.chat_completion.protocol import ChatCompletionRequest
+from vllm.entrypoints.openai.chat_completion.protocol import (
+    ChatCompletionRequest,
+    ChatCompletionToolsParam,
+)
 from vllm.entrypoints.openai.engine.protocol import (
     DeltaMessage,
     ExtractedToolCallInformation,
@@ -30,6 +35,8 @@ from vllm.utils.import_utils import import_from_path
 
 logger = init_logger(__name__)
 
+Tool: TypeAlias = ChatCompletionToolsParam | ResponsesTool
+
 
 class ToolParser:
     """
@@ -38,7 +45,11 @@ class ToolParser:
     derived classes.
     """
 
-    def __init__(self, tokenizer: TokenizerLike):
+    def __init__(
+        self,
+        tokenizer: TokenizerLike,
+        tools: list[Tool] | None = None,
+    ):
         self.prev_tool_call_arr: list[dict] = []
         # the index of the tool call that is currently being parsed
         self.current_tool_id: int = -1
@@ -46,6 +57,7 @@ class ToolParser:
         self.streamed_args_for_tool: list[str] = []
 
         self.model_tokenizer = tokenizer
+        self.tools = tools
 
     @cached_property
     def vocab(self) -> dict[str, int]:

vllm/tool_parsers/deepseekv31_tool_parser.py

@@ -19,14 +19,14 @@ from vllm.entrypoints.openai.engine.protocol import (
 )
 from vllm.logger import init_logger
 from vllm.tokenizers import TokenizerLike
-from vllm.tool_parsers.abstract_tool_parser import ToolParser
+from vllm.tool_parsers.abstract_tool_parser import Tool, ToolParser
 
 logger = init_logger(__name__)
 
 
 class DeepSeekV31ToolParser(ToolParser):
-    def __init__(self, tokenizer: TokenizerLike):
-        super().__init__(tokenizer)
+    def __init__(self, tokenizer: TokenizerLike, tools: list[Tool] | None = None):
+        super().__init__(tokenizer, tools)
 
         self.current_tool_name_sent: bool = False
         self.prev_tool_call_arr: list[dict] = []

vllm/tool_parsers/deepseekv32_tool_parser.py

@@ -22,6 +22,7 @@ from vllm.entrypoints.openai.engine.protocol import (
 from vllm.logger import init_logger
 from vllm.tokenizers import TokenizerLike
 from vllm.tool_parsers.abstract_tool_parser import (
+    Tool,
     ToolParser,
 )
 
@@ -43,8 +44,8 @@ class DeepSeekV32ToolParser(ToolParser):
     </｜DSML｜function_calls>
     """
 
-    def __init__(self, tokenizer: TokenizerLike):
-        super().__init__(tokenizer)
+    def __init__(self, tokenizer: TokenizerLike, tools: list[Tool] | None = None):
+        super().__init__(tokenizer, tools)
 
         self.prev_tool_call_arr: list[dict] = []
 

vllm/tool_parsers/deepseekv3_tool_parser.py

@@ -20,6 +20,7 @@ from vllm.entrypoints.openai.engine.protocol import (
 from vllm.logger import init_logger
 from vllm.tokenizers import TokenizerLike
 from vllm.tool_parsers.abstract_tool_parser import (
+    Tool,
     ToolParser,
 )
 
@@ -27,8 +28,8 @@ logger = init_logger(__name__)
 
 
 class DeepSeekV3ToolParser(ToolParser):
-    def __init__(self, tokenizer: TokenizerLike):
-        super().__init__(tokenizer)
+    def __init__(self, tokenizer: TokenizerLike, tools: list[Tool] | None = None):
+        super().__init__(tokenizer, tools)
 
         self.current_tool_name_sent: bool = False
         self.prev_tool_call_arr: list[dict] = []

vllm/tool_parsers/ernie45_tool_parser.py

@@ -20,6 +20,7 @@ from vllm.entrypoints.openai.engine.protocol import (
 from vllm.logger import init_logger
 from vllm.tokenizers import TokenizerLike
 from vllm.tool_parsers.abstract_tool_parser import (
+    Tool,
     ToolParser,
 )
 
@@ -27,12 +28,12 @@ logger = init_logger(__name__)
 
 
 class Ernie45ToolParser(ToolParser):
-    def __init__(self, tokenizer: TokenizerLike):
+    def __init__(self, tokenizer: TokenizerLike, tools: list[Tool] | None = None):
         """
         Ernie thinking model format:
         abc\n</think>\n\n\n<tool_call>\ndef\n</tool_call>\n
         """
-        super().__init__(tokenizer)
+        super().__init__(tokenizer, tools)
         self.current_tool_name_sent = False
         self.prev_tool_call_arr: list[dict] = []
         self.current_tool_id = -1

vllm/tool_parsers/functiongemma_tool_parser.py

@@ -20,7 +20,7 @@ from vllm.entrypoints.openai.engine.protocol import (
 )
 from vllm.logger import init_logger
 from vllm.tokenizers import TokenizerLike
-from vllm.tool_parsers.abstract_tool_parser import ToolParser
+from vllm.tool_parsers.abstract_tool_parser import Tool, ToolParser
 
 logger = init_logger(__name__)
 
@@ -33,8 +33,8 @@ class FunctionGemmaToolParser(ToolParser):
     <start_function_call>call:func_name{param:<escape>value<escape>}<end_function_call>
     """
 
-    def __init__(self, tokenizer: TokenizerLike):
-        super().__init__(tokenizer)
+    def __init__(self, tokenizer: TokenizerLike, tools: list[Tool] | None = None):
+        super().__init__(tokenizer, tools)
 
         # Streaming state
         self.current_tool_name_sent: bool = False

vllm/tool_parsers/gigachat3_tool_parser.py

@@ -20,7 +20,7 @@ from vllm.entrypoints.openai.engine.protocol import (
 )
 from vllm.logger import init_logger
 from vllm.tokenizers import TokenizerLike
-from vllm.tool_parsers.abstract_tool_parser import ToolParser
+from vllm.tool_parsers.abstract_tool_parser import Tool, ToolParser
 
 logger = init_logger(__name__)
 
@@ -46,8 +46,8 @@ ARGS_REGEX = re.compile(
 
 
 class GigaChat3ToolParser(ToolParser):
-    def __init__(self, tokenizer: TokenizerLike):
-        super().__init__(tokenizer)
+    def __init__(self, tokenizer: TokenizerLike, tools: list[Tool] | None = None):
+        super().__init__(tokenizer, tools)
         self.tool_started: bool = False
         self.tool_name_sent: bool = False
         self.tool_id: str | None = None

vllm/tool_parsers/glm47_moe_tool_parser.py

@@ -16,14 +16,15 @@ import regex as re
 
 from vllm.logger import init_logger
 from vllm.tokenizers import TokenizerLike
+from vllm.tool_parsers.abstract_tool_parser import Tool
 from vllm.tool_parsers.glm4_moe_tool_parser import Glm4MoeModelToolParser
 
 logger = init_logger(__name__)
 
 
 class Glm47MoeModelToolParser(Glm4MoeModelToolParser):
-    def __init__(self, tokenizer: TokenizerLike):
-        super().__init__(tokenizer)
+    def __init__(self, tokenizer: TokenizerLike, tools: list[Tool] | None = None):
+        super().__init__(tokenizer, tools)
         # GLM-4.7 format: <tool_call>func_name[<arg_key>...]*</tool_call>
         # The function name can be followed by a newline, whitespace, or
         # directly by <arg_key> tags (no separator).  The arg section is

vllm/tool_parsers/glm4_moe_tool_parser.py

@@ -21,7 +21,6 @@ import regex as re
 from vllm.entrypoints.chat_utils import make_tool_call_id
 from vllm.entrypoints.openai.chat_completion.protocol import (
     ChatCompletionRequest,
-    ChatCompletionToolsParam,
 )
 from vllm.entrypoints.openai.engine.protocol import (
     DeltaFunctionCall,
@@ -34,6 +33,7 @@ from vllm.entrypoints.openai.engine.protocol import (
 from vllm.logger import init_logger
 from vllm.tokenizers import TokenizerLike
 from vllm.tool_parsers.abstract_tool_parser import (
+    Tool,
     ToolParser,
 )
 
@@ -48,8 +48,8 @@ class Glm4MoeModelToolParser(ToolParser):
     rather than waiting for the complete </arg_value> tag.
     """
 
-    def __init__(self, tokenizer: TokenizerLike):
-        super().__init__(tokenizer)
+    def __init__(self, tokenizer: TokenizerLike, tools: list[Tool] | None = None):
+        super().__init__(tokenizer, tools)
         # Stateful streaming fields
         self.current_tool_name_sent: bool = False
         self.prev_tool_call_arr: list[dict[str, Any]] = []
@@ -122,7 +122,7 @@ class Glm4MoeModelToolParser(ToolParser):
     def _is_string_type(
         tool_name: str,
         arg_name: str,
-        tools: list[ChatCompletionToolsParam] | None,
+        tools: list[Tool] | None,
     ) -> bool:
         if tools is None:
             return False

vllm/tool_parsers/granite4_tool_parser.py

@@ -22,6 +22,7 @@ from vllm.entrypoints.openai.engine.protocol import (
 from vllm.logger import init_logger
 from vllm.tokenizers import TokenizerLike
 from vllm.tool_parsers.abstract_tool_parser import (
+    Tool,
     ToolParser,
 )
 
@@ -43,8 +44,8 @@ FuncT = TypeVar("FuncT", bound=_FunctionCallCtor)
 
 
 class Granite4ToolParser(ToolParser):
-    def __init__(self, tokenizer: TokenizerLike):
-        super().__init__(tokenizer)
+    def __init__(self, tokenizer: TokenizerLike, tools: list[Tool] | None = None):
+        super().__init__(tokenizer, tools)
 
         self.prev_tool_call_arr: list[dict] = []
         self.current_tool_id: int = -1

vllm/tool_parsers/granite_20b_fc_tool_parser.py

@@ -24,6 +24,7 @@ from vllm.entrypoints.openai.engine.protocol import (
 from vllm.logger import init_logger
 from vllm.tokenizers import TokenizerLike
 from vllm.tool_parsers.abstract_tool_parser import (
+    Tool,
     ToolParser,
 )
 from vllm.tool_parsers.utils import (
@@ -46,8 +47,8 @@ class Granite20bFCToolParser(ToolParser):
     are all set
     """
 
-    def __init__(self, tokenizer: TokenizerLike):
-        super().__init__(tokenizer)
+    def __init__(self, tokenizer: TokenizerLike, tools: list[Tool] | None = None):
+        super().__init__(tokenizer, tools)
 
         self.bot_token = "<function_call>"
         self.tool_start_token = self.bot_token

vllm/tool_parsers/granite_tool_parser.py

@@ -22,6 +22,7 @@ from vllm.entrypoints.openai.engine.protocol import (
 from vllm.logger import init_logger
 from vllm.tokenizers import TokenizerLike
 from vllm.tool_parsers.abstract_tool_parser import (
+    Tool,
     ToolParser,
 )
 from vllm.tool_parsers.utils import (
@@ -44,8 +45,8 @@ class GraniteToolParser(ToolParser):
     are all set
     """
 
-    def __init__(self, tokenizer: TokenizerLike):
-        super().__init__(tokenizer)
+    def __init__(self, tokenizer: TokenizerLike, tools: list[Tool] | None = None):
+        super().__init__(tokenizer, tools)
         # for granite 3.0, the token `<|tool_call|>`
         self.bot_token = "<|tool_call|>"
         # for granite 3.1, the string `<tool_call>`

vllm/tool_parsers/hermes_tool_parser.py

@@ -23,6 +23,7 @@ from vllm.entrypoints.openai.engine.protocol import (
 from vllm.logger import init_logger
 from vllm.tokenizers import TokenizerLike
 from vllm.tool_parsers.abstract_tool_parser import (
+    Tool,
     ToolParser,
 )
 from vllm.utils.mistral import is_mistral_tokenizer
@@ -31,8 +32,8 @@ logger = init_logger(__name__)
 
 
 class Hermes2ProToolParser(ToolParser):
-    def __init__(self, tokenizer: TokenizerLike):
-        super().__init__(tokenizer)
+    def __init__(self, tokenizer: TokenizerLike, tools: list[Tool] | None = None):
+        super().__init__(tokenizer, tools)
 
         if is_mistral_tokenizer(tokenizer):
             logger.error("Detected Mistral tokenizer when using a Hermes model")

vllm/tool_parsers/hunyuan_a13b_tool_parser.py

@@ -22,6 +22,7 @@ from vllm.entrypoints.openai.engine.protocol import (
 from vllm.logger import init_logger
 from vllm.tokenizers import TokenizerLike
 from vllm.tool_parsers.abstract_tool_parser import (
+    Tool,
     ToolParser,
 )
 from vllm.tool_parsers.utils import consume_space
@@ -31,8 +32,8 @@ logger = init_logger(__name__)
 
 
 class HunyuanA13BToolParser(ToolParser):
-    def __init__(self, tokenizer: TokenizerLike):
-        super().__init__(tokenizer)
+    def __init__(self, tokenizer: TokenizerLike, tools: list[Tool] | None = None):
+        super().__init__(tokenizer, tools)
 
         # Initialize state for streaming mode
         self.prev_tool_calls: list[dict] = []

vllm/tool_parsers/internlm2_tool_parser.py

@@ -22,6 +22,7 @@ from vllm.entrypoints.openai.engine.protocol import (
 from vllm.logger import init_logger
 from vllm.tokenizers import TokenizerLike
 from vllm.tool_parsers.abstract_tool_parser import (
+    Tool,
     ToolParser,
 )
 from vllm.tool_parsers.utils import extract_intermediate_diff
@@ -30,8 +31,8 @@ logger = init_logger(__name__)
 
 
 class Internlm2ToolParser(ToolParser):
-    def __init__(self, tokenizer: TokenizerLike):
-        super().__init__(tokenizer)
+    def __init__(self, tokenizer: TokenizerLike, tools: list[Tool] | None = None):
+        super().__init__(tokenizer, tools)
         self.position = 0
 
     def adjust_request(self, request: ChatCompletionRequest) -> ChatCompletionRequest:

vllm/tool_parsers/jamba_tool_parser.py

@@ -22,7 +22,7 @@ from vllm.entrypoints.openai.engine.protocol import (
 )
 from vllm.logger import init_logger
 from vllm.tokenizers import TokenizerLike
-from vllm.tool_parsers import ToolParser
+from vllm.tool_parsers.abstract_tool_parser import Tool, ToolParser
 from vllm.tool_parsers.utils import extract_intermediate_diff
 from vllm.utils.mistral import is_mistral_tokenizer
 
@@ -30,8 +30,8 @@ logger = init_logger(__name__)
 
 
 class JambaToolParser(ToolParser):
-    def __init__(self, tokenizer: TokenizerLike):
-        super().__init__(tokenizer)
+    def __init__(self, tokenizer: TokenizerLike, tools: list[Tool] | None = None):
+        super().__init__(tokenizer, tools)
 
         if is_mistral_tokenizer(self.model_tokenizer):
             raise ValueError(

vllm/tool_parsers/kimi_k2_tool_parser.py

@@ -20,6 +20,7 @@ from vllm.entrypoints.openai.engine.protocol import (
 from vllm.logger import init_logger
 from vllm.tokenizers import TokenizerLike
 from vllm.tool_parsers.abstract_tool_parser import (
+    Tool,
     ToolParser,
 )
 
@@ -27,8 +28,8 @@ logger = init_logger(__name__)
 
 
 class KimiK2ToolParser(ToolParser):
-    def __init__(self, tokenizer: TokenizerLike):
-        super().__init__(tokenizer)
+    def __init__(self, tokenizer: TokenizerLike, tools: list[Tool] | None = None):
+        super().__init__(tokenizer, tools)
         self.current_tool_name_sent: bool = False
         self.prev_tool_call_arr: list[dict] = []
         self.current_tool_id: int = -1

vllm/tool_parsers/llama4_pythonic_tool_parser.py

@@ -17,6 +17,7 @@ from vllm.entrypoints.openai.engine.protocol import (
 )
 from vllm.logger import init_logger
 from vllm.tool_parsers.abstract_tool_parser import (
+    Tool,
     ToolParser,
 )
 from vllm.tool_parsers.utils import (
@@ -47,8 +48,12 @@ class Llama4PythonicToolParser(ToolParser):
         re.DOTALL,
     )
 
-    def __init__(self, tokenizer: PreTrainedTokenizerBase):
-        super().__init__(tokenizer)
+    def __init__(
+        self,
+        tokenizer: PreTrainedTokenizerBase,
+        tools: list[Tool] | None = None,
+    ):
+        super().__init__(tokenizer, tools)
 
     # Rename for readability. This is NOT a tool id.
     @property

vllm/tool_parsers/llama_tool_parser.py

@@ -24,6 +24,7 @@ from vllm.entrypoints.openai.engine.protocol import (
 )
 from vllm.logger import init_logger
 from vllm.tool_parsers.abstract_tool_parser import (
+    Tool,
     ToolParser,
 )
 from vllm.tool_parsers.utils import (
@@ -44,8 +45,12 @@ class Llama3JsonToolParser(ToolParser):
     llama4_json are set.
     """
 
-    def __init__(self, tokenizer: PreTrainedTokenizerBase):
-        super().__init__(tokenizer)
+    def __init__(
+        self,
+        tokenizer: PreTrainedTokenizerBase,
+        tools: list[Tool] | None = None,
+    ):
+        super().__init__(tokenizer, tools)
 
         # initialize properties used for state when parsing tool calls in
         # streaming mode

vllm/tool_parsers/longcat_tool_parser.py

@@ -4,12 +4,13 @@
 import regex as re
 
 from vllm.tokenizers import TokenizerLike
+from vllm.tool_parsers.abstract_tool_parser import Tool
 from vllm.tool_parsers.hermes_tool_parser import Hermes2ProToolParser
 
 
 class LongcatFlashToolParser(Hermes2ProToolParser):
-    def __init__(self, tokenizer: TokenizerLike):
-        super().__init__(tokenizer)
+    def __init__(self, tokenizer: TokenizerLike, tools: list[Tool] | None = None):
+        super().__init__(tokenizer, tools)
 
         self.tool_call_start_token: str = "<longcat_tool_call>"
         self.tool_call_end_token: str = "</longcat_tool_call>"

vllm/tool_parsers/minimax_m2_tool_parser.py

@@ -22,6 +22,7 @@ from vllm.entrypoints.openai.engine.protocol import (
 from vllm.logger import init_logger
 from vllm.tokenizers import TokenizerLike
 from vllm.tool_parsers.abstract_tool_parser import (
+    Tool,
     ToolParser,
 )
 
@@ -29,8 +30,8 @@ logger = init_logger(__name__)
 
 
 class MinimaxM2ToolParser(ToolParser):
-    def __init__(self, tokenizer: TokenizerLike):
-        super().__init__(tokenizer)
+    def __init__(self, tokenizer: TokenizerLike, tools: list[Tool] | None = None):
+        super().__init__(tokenizer, tools)
 
         self.prev_tool_call_arr: list[dict] = []
 

vllm/tool_parsers/minimax_tool_parser.py

@@ -22,6 +22,7 @@ from vllm.entrypoints.openai.engine.protocol import (
 from vllm.logger import init_logger
 from vllm.tokenizers import TokenizerLike
 from vllm.tool_parsers.abstract_tool_parser import (
+    Tool,
     ToolParser,
 )
 from vllm.tool_parsers.utils import extract_intermediate_diff
@@ -30,8 +31,8 @@ logger = init_logger(__name__)
 
 
 class MinimaxToolParser(ToolParser):
-    def __init__(self, tokenizer: TokenizerLike):
-        super().__init__(tokenizer)
+    def __init__(self, tokenizer: TokenizerLike, tools: list[Tool] | None = None):
+        super().__init__(tokenizer, tools)
 
         # Initialize streaming state for tracking tool call progress
         self.streaming_state: dict[str, Any] = {

vllm/tool_parsers/mistral_tool_parser.py

@@ -26,6 +26,7 @@ from vllm.entrypoints.openai.engine.protocol import (
 from vllm.logger import init_logger
 from vllm.tokenizers import TokenizerLike
 from vllm.tool_parsers.abstract_tool_parser import (
+    Tool,
     ToolParser,
 )
 from vllm.utils.mistral import is_mistral_tokenizer
@@ -78,8 +79,8 @@ class MistralToolParser(ToolParser):
     Used when --enable-auto-tool-choice --tool-call-parser mistral are all set
     """
 
-    def __init__(self, tokenizer: TokenizerLike):
-        super().__init__(tokenizer)
+    def __init__(self, tokenizer: TokenizerLike, tools: list[Tool] | None = None):
+        super().__init__(tokenizer, tools)
 
         if not is_mistral_tokenizer(self.model_tokenizer):
             logger.info("Non-Mistral tokenizer detected when using a Mistral model...")

vllm/tool_parsers/olmo3_tool_parser.py

@@ -17,6 +17,7 @@ from vllm.entrypoints.openai.engine.protocol import (
 )
 from vllm.logger import init_logger
 from vllm.tool_parsers.abstract_tool_parser import (
+    Tool,
     ToolParser,
 )
 from vllm.tool_parsers.utils import (
@@ -51,8 +52,12 @@ class Olmo3PythonicToolParser(ToolParser):
         re.DOTALL,
     )
 
-    def __init__(self, tokenizer: PreTrainedTokenizerBase):
-        super().__init__(tokenizer)
+    def __init__(
+        self,
+        tokenizer: PreTrainedTokenizerBase,
+        tools: list[Tool] | None = None,
+    ):
+        super().__init__(tokenizer, tools)
 
     # Rename for readability. This is NOT a tool id.
     @property

vllm/tool_parsers/openai_tool_parser.py

@@ -16,6 +16,7 @@ from vllm.entrypoints.openai.engine.protocol import (
 from vllm.entrypoints.openai.parser.harmony_utils import parse_output_into_messages
 from vllm.logger import init_logger
 from vllm.tool_parsers.abstract_tool_parser import (
+    Tool,
     ToolParser,
 )
 
@@ -28,8 +29,8 @@ logger = init_logger(__name__)
 
 
 class OpenAIToolParser(ToolParser):
-    def __init__(self, tokenizer: "TokenizerLike"):
-        super().__init__(tokenizer)
+    def __init__(self, tokenizer: "TokenizerLike", tools: list[Tool] | None = None):
+        super().__init__(tokenizer, tools)
 
     def extract_tool_calls(
         self,

vllm/tool_parsers/phi4mini_tool_parser.py

@@ -20,6 +20,7 @@ from vllm.entrypoints.openai.engine.protocol import (
 )
 from vllm.logger import init_logger
 from vllm.tool_parsers.abstract_tool_parser import (
+    Tool,
     ToolParser,
 )
 
@@ -35,8 +36,12 @@ class Phi4MiniJsonToolParser(ToolParser):
     are all set
     """
 
-    def __init__(self, tokenizer: PreTrainedTokenizerBase) -> None:
-        super().__init__(tokenizer)
+    def __init__(
+        self,
+        tokenizer: PreTrainedTokenizerBase,
+        tools: list[Tool] | None = None,
+    ) -> None:
+        super().__init__(tokenizer, tools)
 
         # initialize properties used for state when parsing tool calls in
         # streaming mode

vllm/tool_parsers/pythonic_tool_parser.py

@@ -17,6 +17,7 @@ from vllm.entrypoints.openai.engine.protocol import (
 )
 from vllm.logger import init_logger
 from vllm.tool_parsers.abstract_tool_parser import (
+    Tool,
     ToolParser,
 )
 from vllm.tool_parsers.utils import (
@@ -49,8 +50,12 @@ class PythonicToolParser(ToolParser):
         re.DOTALL,
     )
 
-    def __init__(self, tokenizer: PreTrainedTokenizerBase):
-        super().__init__(tokenizer)
+    def __init__(
+        self,
+        tokenizer: PreTrainedTokenizerBase,
+        tools: list[Tool] | None = None,
+    ):
+        super().__init__(tokenizer, tools)
 
     # Rename for readability. This is NOT a tool id.
     @property

vllm/tool_parsers/qwen3coder_tool_parser.py

@@ -10,7 +10,6 @@ import regex as re
 
 from vllm.entrypoints.openai.chat_completion.protocol import (
     ChatCompletionRequest,
-    ChatCompletionToolsParam,
 )
 from vllm.entrypoints.openai.engine.protocol import (
     DeltaFunctionCall,
@@ -23,6 +22,7 @@ from vllm.entrypoints.openai.engine.protocol import (
 from vllm.logger import init_logger
 from vllm.tokenizers import TokenizerLike
 from vllm.tool_parsers.abstract_tool_parser import (
+    Tool,
     ToolParser,
 )
 
@@ -30,8 +30,8 @@ logger = init_logger(__name__)
 
 
 class Qwen3CoderToolParser(ToolParser):
-    def __init__(self, tokenizer: TokenizerLike):
-        super().__init__(tokenizer)
+    def __init__(self, tokenizer: TokenizerLike, tools: list[Tool] | None = None):
+        super().__init__(tokenizer, tools)
 
         self.current_tool_name_sent: bool = False
         self.prev_tool_call_arr: list[dict] = []
@@ -109,9 +109,7 @@ class Qwen3CoderToolParser(ToolParser):
         self.accumulated_params = {}
         self.streaming_request = None
 
-    def _get_arguments_config(
-        self, func_name: str, tools: list[ChatCompletionToolsParam] | None
-    ) -> dict:
+    def _get_arguments_config(self, func_name: str, tools: list[Tool] | None) -> dict:
         """Extract argument configuration for a function."""
         if tools is None:
             return {}
@@ -246,7 +244,7 @@ class Qwen3CoderToolParser(ToolParser):
             return param_value
 
     def _parse_xml_function_call(
-        self, function_call_str: str, tools: list[ChatCompletionToolsParam] | None
+        self, function_call_str: str, tools: list[Tool] | None
     ) -> ToolCall | None:
         # Extract function name
         end_index = function_call_str.find(">")

vllm/tool_parsers/qwen3xml_tool_parser.py

@@ -11,7 +11,6 @@ import regex as re
 from vllm.entrypoints.chat_utils import make_tool_call_id
 from vllm.entrypoints.openai.chat_completion.protocol import (
     ChatCompletionRequest,
-    ChatCompletionToolsParam,
 )
 from vllm.entrypoints.openai.engine.protocol import (
     DeltaFunctionCall,
@@ -24,6 +23,7 @@ from vllm.entrypoints.openai.engine.protocol import (
 from vllm.logger import init_logger
 from vllm.tokenizers import TokenizerLike
 from vllm.tool_parsers.abstract_tool_parser import (
+    Tool,
     ToolParser,
 )
 
@@ -40,7 +40,7 @@ class StreamingXMLToolCallParser:
         self.reset_streaming_state()
 
         # Tool configuration information
-        self.tools: list[ChatCompletionToolsParam] | None = None
+        self.tools: list[Tool] | None = None
         self.tool_call_start_token: str = "<tool_call>"
         self.tool_call_end_token: str = "</tool_call>"
         self.function_start_token: str = "<function="
@@ -961,7 +961,7 @@ class StreamingXMLToolCallParser:
         self.parser.EndElementHandler = self._end_element
         self.parser.CharacterDataHandler = self._char_data
 
-    def set_tools(self, tools: list[ChatCompletionToolsParam] | None):
+    def set_tools(self, tools: list[Tool] | None):
         """Set tool configuration information"""
         self.tools = tools
 
@@ -1167,8 +1167,8 @@ class StreamingXMLToolCallParser:
 
 
 class Qwen3XMLToolParser(ToolParser):
-    def __init__(self, tokenizer: TokenizerLike):
-        super().__init__(tokenizer)
+    def __init__(self, tokenizer: TokenizerLike, tools: list[Tool] | None = None):
+        super().__init__(tokenizer, tools)
         self.parser = StreamingXMLToolCallParser()
 
         # Add missing attributes for compatibility with serving_chat.py

vllm/tool_parsers/seed_oss_tool_parser.py

@@ -13,7 +13,6 @@ import regex as re
 
 from vllm.entrypoints.openai.chat_completion.protocol import (
     ChatCompletionRequest,
-    ChatCompletionToolsParam,
 )
 from vllm.entrypoints.openai.engine.protocol import (
     DeltaFunctionCall,
@@ -26,6 +25,7 @@ from vllm.entrypoints.openai.engine.protocol import (
 from vllm.logger import init_logger
 from vllm.tokenizers import TokenizerLike
 from vllm.tool_parsers.abstract_tool_parser import (
+    Tool,
     ToolParser,
 )
 
@@ -36,8 +36,8 @@ class SeedOssToolParser(ToolParser):
     TOOL_CALL_START = "<seed:tool_call>"
     TOOL_CALL_END = "</seed:tool_call>"
 
-    def __init__(self, tokenizer: TokenizerLike):
-        super().__init__(tokenizer)
+    def __init__(self, tokenizer: TokenizerLike, tools: list[Tool] | None = None):
+        super().__init__(tokenizer, tools)
 
         # --- streaming state ---
         self._reset_streaming_state()
@@ -109,7 +109,7 @@ class SeedOssToolParser(ToolParser):
         self.json_closed = False
 
     def _parse_xml_function_call(
-        self, function_call_str: str, tools: list[ChatCompletionToolsParam] | None
+        self, function_call_str: str, tools: list[Tool] | None
     ) -> ToolCall | None:
         def get_arguments_config(func_name: str) -> dict:
             if tools is None:

vllm/tool_parsers/step3_tool_parser.py

@@ -22,6 +22,7 @@ from vllm.entrypoints.openai.engine.protocol import (
 from vllm.logger import init_logger
 from vllm.tokenizers import TokenizerLike
 from vllm.tool_parsers.abstract_tool_parser import (
+    Tool,
     ToolParser,
 )
 from vllm.utils import random_uuid
@@ -43,8 +44,8 @@ class Step3ToolParser(ToolParser):
     TOOL_SEP = "<｜tool_sep｜>"
     SPECIAL_TOKENS = [TOOL_CALLS_BEGIN, TOOL_CALLS_END, TOOL_CALL_BEGIN, TOOL_CALL_END]
 
-    def __init__(self, tokenizer: TokenizerLike):
-        super().__init__(tokenizer)
+    def __init__(self, tokenizer: TokenizerLike, tools: list[Tool] | None = None):
+        super().__init__(tokenizer, tools)
         self.position = 0
         # Explicit state flags for robust streaming
         self.tool_block_started = False

vllm/tool_parsers/step3p5_tool_parser.py

@@ -11,7 +11,6 @@ import regex as re
 from vllm.entrypoints.chat_utils import make_tool_call_id
 from vllm.entrypoints.openai.chat_completion.protocol import (
     ChatCompletionRequest,
-    ChatCompletionToolsParam,
 )
 from vllm.entrypoints.openai.engine.protocol import (
     DeltaFunctionCall,
@@ -23,7 +22,7 @@ from vllm.entrypoints.openai.engine.protocol import (
 )
 from vllm.logger import init_logger
 from vllm.tokenizers import TokenizerLike
-from vllm.tool_parsers.abstract_tool_parser import ToolParser
+from vllm.tool_parsers.abstract_tool_parser import Tool, ToolParser
 
 logger = init_logger(__name__)
 
@@ -38,7 +37,7 @@ class StreamingXMLToolCallParser:
         self.reset_streaming_state()
 
         # Tool configuration information
-        self.tools: list[ChatCompletionToolsParam] | None = None
+        self.tools: list[Tool] | None = None
         self.tool_call_start_token: str = "<tool_call>"
         self.tool_call_end_token: str = "</tool_call>"
         self.function_start_token: str = "<function="
@@ -1161,7 +1160,7 @@ class StreamingXMLToolCallParser:
         self.parser.EndElementHandler = self._end_element
         self.parser.CharacterDataHandler = self._char_data
 
-    def set_tools(self, tools: list[ChatCompletionToolsParam] | None):
+    def set_tools(self, tools: list[Tool] | None):
         """Set tool configuration information"""
         self.tools = tools
 
@@ -1365,8 +1364,8 @@ class StreamingXMLToolCallParser:
 
 
 class Step3p5ToolParser(ToolParser):
-    def __init__(self, tokenizer: TokenizerLike):
-        super().__init__(tokenizer)
+    def __init__(self, tokenizer: TokenizerLike, tools: list[Tool] | None = None):
+        super().__init__(tokenizer, tools)
         self.parser = StreamingXMLToolCallParser()
 
         # Add missing attributes for compatibility with serving_chat.py

vllm/tool_parsers/xlam_tool_parser.py

@@ -6,6 +6,7 @@ from collections.abc import Sequence
 from typing import Any, Optional, Union
 
 import regex as re
+
 from vllm.entrypoints.openai.chat_completion.protocol import (
     ChatCompletionRequest,
 )
@@ -19,6 +20,7 @@ from vllm.entrypoints.openai.engine.protocol import (
     ToolCall,
 )
 from vllm.tool_parsers.abstract_tool_parser import (
+    Tool,
     ToolParser,
 )
 from vllm.logger import init_logger
@@ -29,8 +31,8 @@ logger = init_logger(__name__)
 
 
 class xLAMToolParser(ToolParser):
-    def __init__(self, tokenizer: TokenizerLike):
-        super().__init__(tokenizer)
+    def __init__(self, tokenizer: TokenizerLike, tools: list[Tool] | None = None):
+        super().__init__(tokenizer, tools)
 
         # Initialize state for streaming mode
         self.prev_tool_calls: list[dict] = []