PoC: InferenceClient is also a MCPClient (#2986)

* Add extra dependency

* PoC: `InferenceClient` is also a `MCPClient`

* [using Claude] change the code to make MCPClient inherit from AsyncInferenceClient

* Update mcp_client.py

Co-authored-by: Mishig <dmishig@gmail.com>

* mcp_client: Support multiple servers (#2987)

cc @mishig25

* Revert "[using Claude] change the code to make MCPClient inherit from AsyncInferenceClient"

This reverts commit 2c7329c.

* `add_mcp_server`: the env should not be hardcoded here

cc @Wauplin

* Handle the "no tool call" case

* Update setup.py

* Async mcp client + example + code quality

* docstring

* accept ChatCompletionInputMessage as input

* lazy loading

* style

* better type

* no need mcp for dev

* code quality on Python 3.8

* address feedback

* address feedback

* do not close client inside of �[200~process_single_turn_with_tools~

* docstring, no more warning, garbage collection

* docs

---------

Co-authored-by: Mishig <dmishig@gmail.com>
Co-authored-by: Lucain Pouget <lucainp@gmail.com>
Co-authored-by: Lucain <lucain@huggingface.co>

Author: 3 people

setup.py

@@ -64,6 +64,10 @@ def get_version() -> str:
 
 extras["hf_xet"] = ["hf-xet>=1.1.1,<2.0.0"]
 
+extras["mcp"] = [
+    "mcp>=1.8.0",
+] + extras["inference"]
+
 extras["testing"] = (
     extras["cli"]
     + extras["inference"]

src/huggingface_hub/__init__.py

@@ -440,6 +440,9 @@
         "ZeroShotObjectDetectionOutputElement",
         "ZeroShotObjectDetectionParameters",
     ],
+    "inference._mcp.mcp_client": [
+        "MCPClient",
+    ],
     "inference_api": [
         "InferenceApi",
     ],
@@ -644,6 +647,7 @@
     "InferenceEndpointType",
     "InferenceTimeoutError",
     "KerasModelHubMixin",
+    "MCPClient",
     "ModelCard",
     "ModelCardData",
     "ModelHubMixin",
@@ -1402,6 +1406,7 @@ def __dir__():
         ZeroShotObjectDetectionOutputElement,  # noqa: F401
         ZeroShotObjectDetectionParameters,  # noqa: F401
     )
+    from .inference._mcp.mcp_client import MCPClient  # noqa: F401
     from .inference_api import InferenceApi  # noqa: F401
     from .keras_mixin import (
         KerasModelHubMixin,  # noqa: F401

src/huggingface_hub/inference/_client.py

@@ -66,6 +66,7 @@
     AudioToAudioOutputElement,
     AutomaticSpeechRecognitionOutput,
     ChatCompletionInputGrammarType,
+    ChatCompletionInputMessage,
     ChatCompletionInputStreamOptions,
     ChatCompletionInputTool,
     ChatCompletionInputToolChoiceClass,
@@ -100,7 +101,7 @@
     ZeroShotClassificationOutputElement,
     ZeroShotImageClassificationOutputElement,
 )
-from huggingface_hub.inference._providers import PROVIDER_T, get_provider_helper
+from huggingface_hub.inference._providers import PROVIDER_OR_POLICY_T, get_provider_helper
 from huggingface_hub.utils import build_hf_headers, get_session, hf_raise_for_status
 from huggingface_hub.utils._auth import get_token
 from huggingface_hub.utils._deprecation import _deprecate_method
@@ -164,7 +165,7 @@ def __init__(
         self,
         model: Optional[str] = None,
         *,
-        provider: Union[Literal["auto"], PROVIDER_T, None] = None,
+        provider: Optional[PROVIDER_OR_POLICY_T] = None,
         token: Optional[str] = None,
         timeout: Optional[float] = None,
         headers: Optional[Dict[str, str]] = None,
@@ -446,7 +447,7 @@ def automatic_speech_recognition(
     @overload
     def chat_completion(  # type: ignore
         self,
-        messages: List[Dict],
+        messages: List[Union[Dict, ChatCompletionInputMessage]],
         *,
         model: Optional[str] = None,
         stream: Literal[False] = False,
@@ -472,7 +473,7 @@ def chat_completion(  # type: ignore
     @overload
     def chat_completion(  # type: ignore
         self,
-        messages: List[Dict],
+        messages: List[Union[Dict, ChatCompletionInputMessage]],
         *,
         model: Optional[str] = None,
         stream: Literal[True] = True,
@@ -498,7 +499,7 @@ def chat_completion(  # type: ignore
     @overload
     def chat_completion(
         self,
-        messages: List[Dict],
+        messages: List[Union[Dict, ChatCompletionInputMessage]],
         *,
         model: Optional[str] = None,
         stream: bool = False,
@@ -523,7 +524,7 @@ def chat_completion(
 
     def chat_completion(
         self,
-        messages: List[Dict],
+        messages: List[Union[Dict, ChatCompletionInputMessage]],
         *,
         model: Optional[str] = None,
         stream: bool = False,

src/huggingface_hub/inference/_generated/_async_client.py

@@ -51,6 +51,7 @@
     AudioToAudioOutputElement,
     AutomaticSpeechRecognitionOutput,
     ChatCompletionInputGrammarType,
+    ChatCompletionInputMessage,
     ChatCompletionInputStreamOptions,
     ChatCompletionInputTool,
     ChatCompletionInputToolChoiceClass,
@@ -85,7 +86,7 @@
     ZeroShotClassificationOutputElement,
     ZeroShotImageClassificationOutputElement,
 )
-from huggingface_hub.inference._providers import PROVIDER_T, get_provider_helper
+from huggingface_hub.inference._providers import PROVIDER_OR_POLICY_T, get_provider_helper
 from huggingface_hub.utils import build_hf_headers, get_session, hf_raise_for_status
 from huggingface_hub.utils._auth import get_token
 from huggingface_hub.utils._deprecation import _deprecate_method
@@ -154,7 +155,7 @@ def __init__(
         self,
         model: Optional[str] = None,
         *,
-        provider: Union[Literal["auto"], PROVIDER_T, None] = None,
+        provider: Optional[PROVIDER_OR_POLICY_T] = None,
         token: Optional[str] = None,
         timeout: Optional[float] = None,
         headers: Optional[Dict[str, str]] = None,
@@ -480,7 +481,7 @@ async def automatic_speech_recognition(
     @overload
     async def chat_completion(  # type: ignore
         self,
-        messages: List[Dict],
+        messages: List[Union[Dict, ChatCompletionInputMessage]],
         *,
         model: Optional[str] = None,
         stream: Literal[False] = False,
@@ -506,7 +507,7 @@ async def chat_completion(  # type: ignore
     @overload
     async def chat_completion(  # type: ignore
         self,
-        messages: List[Dict],
+        messages: List[Union[Dict, ChatCompletionInputMessage]],
         *,
         model: Optional[str] = None,
         stream: Literal[True] = True,
@@ -532,7 +533,7 @@ async def chat_completion(  # type: ignore
     @overload
     async def chat_completion(
         self,
-        messages: List[Dict],
+        messages: List[Union[Dict, ChatCompletionInputMessage]],
         *,
         model: Optional[str] = None,
         stream: bool = False,
@@ -557,7 +558,7 @@ async def chat_completion(
 
     async def chat_completion(
         self,
-        messages: List[Dict],
+        messages: List[Union[Dict, ChatCompletionInputMessage]],
         *,
         model: Optional[str] = None,
         stream: bool = False,