integrate connectors api within responses api impl in agentic provider

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

client-sdks/stainless/openapi.yml

@@ -7258,8 +7258,13 @@ components:
           type: string
           title: Server Label
         server_url:
-          type: string
+          anyOf:
-          title: Server Url
+          - type: string
+          - type: 'null'
+        connector_id:
+          anyOf:
+          - type: string
+          - type: 'null'
         headers:
           anyOf:
           - additionalProperties: true
@@ -7292,7 +7297,6 @@ components:
       type: object
       required:
       - server_label
-      - server_url
       title: OpenAIResponseInputToolMCP
       description: Model Context Protocol (MCP) tool configuration for OpenAI response inputs.
     CreateOpenaiResponseRequest:

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

@@ -4039,8 +4039,13 @@ components:
           type: string
           title: Server Label
         server_url:
-          type: string
+          anyOf:
-          title: Server Url
+          - type: string
+          - type: 'null'
+        connector_id:
+          anyOf:
+          - type: string
+          - type: 'null'
         headers:
           anyOf:
           - additionalProperties: true
@@ -4073,7 +4078,6 @@ components:
       type: object
       required:
       - server_label
-      - server_url
       title: OpenAIResponseInputToolMCP
       description: Model Context Protocol (MCP) tool configuration for OpenAI response inputs.
     CreateOpenaiResponseRequest:

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

@@ -3792,8 +3792,13 @@ components:
           type: string
           title: Server Label
         server_url:
-          type: string
+          anyOf:
-          title: Server Url
+          - type: string
+          - type: 'null'
+        connector_id:
+          anyOf:
+          - type: string
+          - type: 'null'
         headers:
           anyOf:
           - additionalProperties: true
@@ -3826,7 +3831,6 @@ components:
       type: object
       required:
       - server_label
-      - server_url
       title: OpenAIResponseInputToolMCP
       description: Model Context Protocol (MCP) tool configuration for OpenAI response inputs.
     OpenAIResponseObject:

docs/static/llama-stack-spec.yaml

@@ -5850,8 +5850,13 @@ components:
           type: string
           title: Server Label
         server_url:
-          type: string
+          anyOf:
-          title: Server Url
+          - type: string
+          - type: 'null'
+        connector_id:
+          anyOf:
+          - type: string
+          - type: 'null'
         headers:
           anyOf:
           - additionalProperties: true
@@ -5884,7 +5889,6 @@ components:
       type: object
       required:
       - server_label
-      - server_url
       title: OpenAIResponseInputToolMCP
       description: Model Context Protocol (MCP) tool configuration for OpenAI response inputs.
     CreateOpenaiResponseRequest:

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

@@ -7258,8 +7258,13 @@ components:
           type: string
           title: Server Label
         server_url:
-          type: string
+          anyOf:
-          title: Server Url
+          - type: string
+          - type: 'null'
+        connector_id:
+          anyOf:
+          - type: string
+          - type: 'null'
         headers:
           anyOf:
           - additionalProperties: true
@@ -7292,7 +7297,6 @@ components:
       type: object
       required:
       - server_label
-      - server_url
       title: OpenAIResponseInputToolMCP
       description: Model Context Protocol (MCP) tool configuration for OpenAI response inputs.
     CreateOpenaiResponseRequest:

src/llama_stack/core/connectors/connectors.py

@@ -14,18 +14,16 @@ from llama_stack.log import get_logger
 from llama_stack.providers.utils.tools.mcp import get_mcp_server_info, list_mcp_tools
 from llama_stack_api import (
     Connector,
+    ConnectorNotFoundError,
     Connectors,
+    ConnectorToolNotFoundError,
     ConnectorType,
     ListConnectorsResponse,
     ListRegistriesResponse,
     ListToolsResponse,
     Registry,
-    ToolDef,
-)
-from llama_stack_api.common.errors import (
-    ConnectorNotFoundError,
-    ConnectorToolNotFoundError,
     RegistryNotFoundError,
+    ToolDef,
 )
 
 logger = get_logger(name=__name__, category="connectors")
@@ -55,6 +53,17 @@ class ConnectorServiceImpl(Connectors):
         self.connectors_map: dict[str, Connector] = {}
         self.registries_map: dict[str, Registry] = {}
 
+    def get_connector_url(self, connector_id: str) -> str | None:
+        """Get the URL of a connector by its ID.
+
+        :param connector_id: The ID of the connector to get the URL for.
+        :returns: The URL of the connector.
+        """
+        connector = self.connectors_map.get(connector_id)
+        if connector is None:
+            return None
+        return connector.url
+
     async def register_connector(
         self,
         url: str,

src/llama_stack/providers/inline/agents/meta_reference/__init__.py

@@ -28,6 +28,7 @@ async def get_provider_impl(
         deps[Api.conversations],
         deps[Api.prompts],
         deps[Api.files],
+        deps[Api.connectors],
         policy,
     )
     await impl.initialize()

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

@@ -11,6 +11,7 @@ from llama_stack.log import get_logger
 from llama_stack.providers.utils.responses.responses_store import ResponsesStore
 from llama_stack_api import (
     Agents,
+    Connectors,
     Conversations,
     Files,
     Inference,
@@ -49,6 +50,7 @@ class MetaReferenceAgentsImpl(Agents):
         conversations_api: Conversations,
         prompts_api: Prompts,
         files_api: Files,
+        connectors_api: Connectors,
         policy: list[AccessRule],
     ):
         self.config = config
@@ -60,6 +62,7 @@ class MetaReferenceAgentsImpl(Agents):
         self.conversations_api = conversations_api
         self.prompts_api = prompts_api
         self.files_api = files_api
+        self.connectors_api = connectors_api
         self.in_memory_store = InmemoryKVStoreImpl()
         self.openai_responses_impl: OpenAIResponsesImpl | None = None
         self.policy = policy
@@ -78,6 +81,7 @@ class MetaReferenceAgentsImpl(Agents):
             conversations_api=self.conversations_api,
             prompts_api=self.prompts_api,
             files_api=self.files_api,
+            connectors_api=self.connectors_api,
         )
 
     async def shutdown(self) -> None:

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

@@ -17,6 +17,7 @@ from llama_stack.providers.utils.responses.responses_store import (
     _OpenAIResponseObjectWithInputAndMessages,
 )
 from llama_stack_api import (
+    Connectors,
     ConversationItem,
     Conversations,
     Files,
@@ -79,6 +80,7 @@ class OpenAIResponsesImpl:
         conversations_api: Conversations,
         prompts_api: Prompts,
         files_api: Files,
+        connectors_api: Connectors,
     ):
         self.inference_api = inference_api
         self.tool_groups_api = tool_groups_api
@@ -94,6 +96,7 @@ class OpenAIResponsesImpl:
         )
         self.prompts_api = prompts_api
         self.files_api = files_api
+        self.connectors_api = connectors_api
 
     async def _prepend_previous_response(
         self,
@@ -494,6 +497,7 @@ class OpenAIResponsesImpl:
             instructions=instructions,
             max_tool_calls=max_tool_calls,
             metadata=metadata,
+            connectors_api=self.connectors_api,
         )
 
         # Stream the response

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

@@ -15,6 +15,7 @@ from llama_stack.providers.utils.inference.prompt_adapter import interleaved_con
 from llama_stack_api import (
     AllowedToolsFilter,
     ApprovalFilter,
+    Connectors,
     Inference,
     MCPListToolsTool,
     ModelNotFoundError,
@@ -121,8 +122,10 @@ class StreamingResponseOrchestrator:
         parallel_tool_calls: bool | None = None,
         max_tool_calls: int | None = None,
         metadata: dict[str, str] | None = None,
+        connectors_api: Connectors | None = None,
     ):
         self.inference_api = inference_api
+        self.connectors_api = connectors_api
         self.ctx = ctx
         self.response_id = response_id
         self.created_at = created_at
@@ -1088,6 +1091,15 @@ class StreamingResponseOrchestrator:
         """Process an MCP tool configuration and emit appropriate streaming events."""
         from llama_stack.providers.utils.tools.mcp import list_mcp_tools
 
+        # Resolve connector_id to server_url if provided
+        if mcp_tool.connector_id and not mcp_tool.server_url:
+            if self.connectors_api is None:
+                raise ValueError("Connectors API not available to resolve connector_id")
+            server_url = self.connectors_api.get_connector_url(mcp_tool.connector_id)
+            if not server_url:
+                raise ValueError(f"Connector {mcp_tool.connector_id} not found")
+            mcp_tool = mcp_tool.model_copy(update={"server_url": server_url})
+
         # Emit mcp_list_tools.in_progress
         self.sequence_number += 1
         yield OpenAIResponseObjectStreamResponseMcpListToolsInProgress(

src/llama_stack/providers/registry/agents.py

@@ -36,6 +36,7 @@ def available_providers() -> list[ProviderSpec]:
                 Api.conversations,
                 Api.prompts,
                 Api.files,
+                Api.connectors,
             ],
             optional_api_dependencies=[
                 Api.safety,

src/llama_stack_api/__init__.py

@@ -46,10 +46,13 @@ from .common.content_types import (
 )
 from .common.errors import (
     ConflictError,
+    ConnectorNotFoundError,
+    ConnectorToolNotFoundError,
     DatasetNotFoundError,
     InvalidConversationIdError,
     ModelNotFoundError,
     ModelTypeError,
+    RegistryNotFoundError,
     ResourceNotFoundError,
     TokenValidationError,
     ToolGroupNotFoundError,
@@ -500,6 +503,8 @@ __all__ = [
     "ConnectorInput",
     "Connectors",
     "ConnectorType",
+    "ConnectorNotFoundError",
+    "ConnectorToolNotFoundError",
     "Conversation",
     "ConversationDeletedResource",
     "ConversationItem",
@@ -775,6 +780,7 @@ __all__ = [
     "Ranker",
     "RegexParserScoringFnParams",
     "Registry",
+    "RegistryNotFoundError",
     "RegistryType",
     "RegistryInput",
     "RemoteProviderConfig",

src/llama_stack_api/openai_responses.py

@@ -5,7 +5,7 @@
 # the root directory of this source tree.
 
 from collections.abc import Sequence
-from typing import Annotated, Any, Literal
+from typing import Annotated, Any, Literal, Self
 
 from pydantic import BaseModel, Field, model_validator
 from typing_extensions import TypedDict
@@ -488,7 +488,8 @@ class OpenAIResponseInputToolMCP(BaseModel):
 
     :param type: Tool type identifier, always "mcp"
     :param server_label: Label to identify this MCP server
-    :param server_url: URL endpoint of the MCP server
+    :param server_url: (Optional) URL endpoint of the MCP server
+    :param connector_id: (Optional) ID of the connector to use for this MCP server
     :param headers: (Optional) HTTP headers to include when connecting to the server
     :param authorization: (Optional) OAuth access token for authenticating with the MCP server
     :param require_approval: Approval requirement for tool calls ("always", "never", or filter)
@@ -497,13 +498,20 @@ class OpenAIResponseInputToolMCP(BaseModel):
 
     type: Literal["mcp"] = "mcp"
     server_label: str
-    server_url: str
+    server_url: str | None = None
+    connector_id: str | None = None
     headers: dict[str, Any] | None = None
     authorization: str | None = Field(default=None, exclude=True)
 
     require_approval: Literal["always"] | Literal["never"] | ApprovalFilter = "never"
     allowed_tools: list[str] | AllowedToolsFilter | None = None
 
+    @model_validator(mode="after")
+    def validate_server_or_connector(self) -> Self:
+        if not self.server_url and not self.connector_id:
+            raise ValueError("Either 'server_url' or 'connector_id' must be provided for MCP tool")
+        return self
+
 
 OpenAIResponseInputTool = Annotated[
     OpenAIResponseInputToolWebSearch