Runtime handling updates (#3451)

## Summary
- Refresh runtime handling around session and tool-call flows.
- Adjust model configuration metadata used by runtime integrations.
- Add focused coverage for the updated behavior.

## Validation
- .venv/bin/python -m pytest tests/model_settings/test_serialization.py
tests/models/test_trace_config.py
tests/mcp/test_streamable_http_client_factory.py
tests/test_run_context_approvals.py
tests/test_run_state.py::TestRunState::test_trace_api_key_serialization_is_opt_in
tests/realtime/test_session.py
- .venv/bin/ruff check <touched files>
- .venv/bin/ruff format --check <touched files>
- git diff --check

Author: adrianbravo-oai

src/agents/extensions/models/any_llm_model.py

@@ -34,6 +34,7 @@
     response_terminal_failure_error,
 )
 from ...models._retry_runtime import should_disable_provider_managed_retries
+from ...models._trace import model_config_for_trace
 from ...models.chatcmpl_converter import Converter
 from ...models.chatcmpl_helpers import HEADERS, HEADERS_OVERRIDE, ChatCmplHelpers
 from ...models.chatcmpl_stream_handler import ChatCmplStreamHandler
@@ -467,12 +468,11 @@ async def _get_response_via_chat(
     ) -> ModelResponse:
         with generation_span(
             model=str(self.model),
-            model_config=model_settings.to_json_dict()
-            | {
-                "base_url": str(self.base_url or ""),
-                "provider": self._provider_name,
-                "model_impl": "any-llm",
-            },
+            model_config=model_config_for_trace(
+                model_settings,
+                base_url=self.base_url or "",
+                extra_config={"provider": self._provider_name, "model_impl": "any-llm"},
+            ),
             disabled=tracing.is_disabled(),
         ) as span_generation:
             response = await self._fetch_chat_response(
@@ -570,12 +570,11 @@ async def _stream_response_via_chat(
     ) -> AsyncIterator[TResponseStreamEvent]:
         with generation_span(
             model=str(self.model),
-            model_config=model_settings.to_json_dict()
-            | {
-                "base_url": str(self.base_url or ""),
-                "provider": self._provider_name,
-                "model_impl": "any-llm",
-            },
+            model_config=model_config_for_trace(
+                model_settings,
+                base_url=self.base_url or "",
+                extra_config={"provider": self._provider_name, "model_impl": "any-llm"},
+            ),
             disabled=tracing.is_disabled(),
         ) as span_generation:
             response, stream = await self._fetch_chat_response(

src/agents/extensions/models/litellm_model.py

@@ -43,6 +43,7 @@
 from ...model_settings import ModelSettings
 from ...models._openai_retry import get_openai_retry_advice
 from ...models._retry_runtime import should_disable_provider_managed_retries
+from ...models._trace import model_config_for_trace
 from ...models.chatcmpl_converter import Converter
 from ...models.chatcmpl_helpers import HEADERS, HEADERS_OVERRIDE, ChatCmplHelpers
 from ...models.chatcmpl_stream_handler import ChatCmplStreamHandler
@@ -213,8 +214,11 @@ async def get_response(
     ) -> ModelResponse:
         with generation_span(
             model=str(self.model),
-            model_config=model_settings.to_json_dict()
-            | {"base_url": str(self.base_url or ""), "model_impl": "litellm"},
+            model_config=model_config_for_trace(
+                model_settings,
+                base_url=self.base_url or "",
+                extra_config={"model_impl": "litellm"},
+            ),
             disabled=tracing.is_disabled(),
         ) as span_generation:
             response = await self._fetch_response(
@@ -327,8 +331,11 @@ async def stream_response(
     ) -> AsyncIterator[TResponseStreamEvent]:
         with generation_span(
             model=str(self.model),
-            model_config=model_settings.to_json_dict()
-            | {"base_url": str(self.base_url or ""), "model_impl": "litellm"},
+            model_config=model_config_for_trace(
+                model_settings,
+                base_url=self.base_url or "",
+                extra_config={"model_impl": "litellm"},
+            ),
             disabled=tracing.is_disabled(),
         ) as span_generation:
             response, stream = await self._fetch_response(

src/agents/mcp/server.py

@@ -99,7 +99,7 @@ def _create_default_streamable_http_client(
     timeout: httpx.Timeout | None = None,
     auth: httpx.Auth | None = None,
 ) -> httpx.AsyncClient:
-    kwargs: dict[str, Any] = {"follow_redirects": True}
+    kwargs: dict[str, Any] = {"follow_redirects": False}
     if timeout is not None:
         kwargs["timeout"] = timeout
     if headers is not None:
@@ -1441,8 +1441,9 @@ def create_streams(
                 auth=self.params.get("auth"),
                 transport_factory=_InitializedNotificationTolerantStreamableHTTPTransport,
             )
-        if httpx_client_factory is not None:
-            kwargs["httpx_client_factory"] = httpx_client_factory
+        kwargs["httpx_client_factory"] = (
+            httpx_client_factory or _create_default_streamable_http_client
+        )
         if "auth" in self.params:
             kwargs["auth"] = self.params["auth"]
         return streamablehttp_client(**kwargs)

src/agents/model_settings.py

@@ -60,6 +60,27 @@ class MCPToolChoice:
 Headers: TypeAlias = Mapping[str, str | Omit]
 ToolChoice: TypeAlias = Literal["auto", "required", "none"] | str | MCPToolChoice | None
 
+_TRACEABLE_MODEL_SETTING_FIELDS = (
+    "temperature",
+    "top_p",
+    "frequency_penalty",
+    "presence_penalty",
+    "tool_choice",
+    "parallel_tool_calls",
+    "truncation",
+    "max_tokens",
+    "reasoning",
+    "verbosity",
+    "metadata",
+    "store",
+    "prompt_cache_retention",
+    "include_usage",
+    "response_include",
+    "top_logprobs",
+    "retry",
+    "context_management",
+)
+
 
 @dataclass
 class ModelSettings:
@@ -199,6 +220,11 @@ def resolve(self, override: ModelSettings | None) -> ModelSettings:
     def to_json_dict(self) -> dict[str, Any]:
         return cast(dict[str, Any], TypeAdapter(ModelSettings).dump_python(self, mode="json"))
 
+    def to_traceable_dict(self) -> dict[str, Any]:
+        """Serialize settings for tracing without provider-specific request extras."""
+        payload = self.to_json_dict()
+        return {key: payload[key] for key in _TRACEABLE_MODEL_SETTING_FIELDS if key in payload}
+
 
 def _merge_retry_settings(
     inherited: ModelRetrySettings | None,

src/agents/models/_trace.py

@@ -0,0 +1,31 @@
+from __future__ import annotations
+
+from typing import Any
+from urllib.parse import urlsplit, urlunsplit
+
+from ..model_settings import ModelSettings
+
+
+def sanitize_url_for_trace(url: object) -> str:
+    """Return a URL safe for tracing by removing auth material and request parameters."""
+    try:
+        parts = urlsplit(str(url))
+    except ValueError:
+        return ""
+
+    netloc = parts.netloc.rsplit("@", 1)[-1]
+    return urlunsplit((parts.scheme, netloc, parts.path, "", ""))
+
+
+def model_config_for_trace(
+    model_settings: ModelSettings,
+    *,
+    base_url: object | None = None,
+    extra_config: dict[str, Any] | None = None,
+) -> dict[str, Any]:
+    config = model_settings.to_traceable_dict()
+    if base_url is not None:
+        config["base_url"] = sanitize_url_for_trace(base_url)
+    if extra_config:
+        config.update(extra_config)
+    return config

src/agents/models/openai_chatcompletions.py

@@ -33,6 +33,7 @@
 from ..util._json import _to_dump_compatible
 from ._openai_retry import get_openai_retry_advice
 from ._retry_runtime import should_disable_provider_managed_retries
+from ._trace import model_config_for_trace
 from .chatcmpl_converter import Converter
 from .chatcmpl_helpers import HEADERS, HEADERS_OVERRIDE, ChatCmplHelpers
 from .chatcmpl_stream_handler import ChatCmplStreamHandler
@@ -147,7 +148,7 @@ async def get_response(
 
         with generation_span(
             model=str(self.model),
-            model_config=model_settings.to_json_dict() | {"base_url": str(self._client.base_url)},
+            model_config=model_config_for_trace(model_settings, base_url=self._client.base_url),
             disabled=tracing.is_disabled(),
         ) as span_generation:
             response = await self._fetch_response(
@@ -281,7 +282,7 @@ async def stream_response(
 
         with generation_span(
             model=str(self.model),
-            model_config=model_settings.to_json_dict() | {"base_url": str(self._client.base_url)},
+            model_config=model_config_for_trace(model_settings, base_url=self._client.base_url),
             disabled=tracing.is_disabled(),
         ) as span_generation:
             response, stream = await self._fetch_response(