fix: pass api_base/api_key to agentic hook follow-up requests

When using third-party Anthropic-compatible providers (e.g. Tencent Cloud)
with custom api_base, agentic hooks (websearch_interception) and
count_tokens handler sent follow-up requests to api.anthropic.com instead
of the configured endpoint, causing 401 authentication errors and
deployment cooldown loops.

- llm_http_handler: _execute_anthropic_agentic_plan now reads api_key and
  api_base from agentic_loop_params and passes them to
  anthropic_messages.acreate()
- token_counter: extract api_base from litellm_params and build the
  count_tokens endpoint URL instead of hardcoding api.anthropic.com
- websearch_interception handler: resolve merge conflict and fix legacy
  _execute_agentic_loop path to also pass api_key/api_base

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

litellm/integrations/websearch_interception/handler.py

@@ -755,10 +755,23 @@ class WebSearchInterceptionLogger(CustomLogger):
         if max_tokens is None:
             max_tokens = cast(int, kwargs.get("max_tokens", 1024))
 
+        # Pass api_key/api_base from agentic_loop_params so follow-up requests
+        # route to the same backend instead of falling back to api.anthropic.com
+        _followup_api_key: Optional[str] = None
+        _followup_api_base: Optional[str] = None
+        if logging_obj is not None:
+            agentic_params = logging_obj.model_call_details.get(
+                "agentic_loop_params", {}
+            )
+            _followup_api_key = agentic_params.get("api_key")
+            _followup_api_base = agentic_params.get("api_base")
+
         return await anthropic_messages.acreate(
             max_tokens=max_tokens,
             messages=request_patch.messages,
             model=request_patch.model or model,
+            api_key=_followup_api_key,
+            api_base=_followup_api_base,
             **optional_params,
             **request_patch.kwargs,
         )

litellm/llms/anthropic/count_tokens/token_counter.py

@@ -54,8 +54,9 @@ class AnthropicTokenCounter(BaseTokenCounter):
         deployment = deployment or {}
         litellm_params = deployment.get("litellm_params", {})
 
-        # Get Anthropic API key from deployment config or environment
+        # Get Anthropic API key and api_base from deployment config or environment
         api_key = litellm_params.get("api_key")
+        api_base = litellm_params.get("api_base")
         if not api_key:
             api_key = os.getenv("ANTHROPIC_API_KEY")
 
@@ -63,11 +64,19 @@ class AnthropicTokenCounter(BaseTokenCounter):
             verbose_logger.warning("No Anthropic API key found for token counting")
             return None
 
+        # Build count_tokens endpoint from api_base if available
+        count_tokens_api_base: Optional[str] = None
+        if api_base:
+            # e.g. https://api.lkeap.cloud.tencent.com/plan/anthropic -> https://api.lkeap.cloud.tencent.com/plan/anthropic/v1/messages/count_tokens
+            base = api_base.rstrip("/")
+            count_tokens_api_base = f"{base}/v1/messages/count_tokens"
+
         try:
             result = await anthropic_count_tokens_handler.handle_count_tokens_request(
                 model=model_to_use,
                 messages=messages,
                 api_key=api_key,
+                api_base=count_tokens_api_base,
                 tools=tools,
                 system=system,
             )

litellm/llms/anthropic/experimental_pass_through/messages/handler.py

@@ -368,10 +368,18 @@ def anthropic_messages_handler(
     # Store agentic loop params in logging object for agentic hooks
     # This provides original request context needed for follow-up calls
     if litellm_logging_obj is not None:
-        litellm_logging_obj.model_call_details["agentic_loop_params"] = {
+        agentic_loop_params: Dict[str, Any] = {
             "model": original_model,
             "custom_llm_provider": custom_llm_provider,
         }
+        # Preserve api_base and api_key so that agentic hooks (e.g. websearch
+        # interception follow-up requests) can reuse the same credentials
+        # instead of falling back to defaults (api.anthropic.com / ANTHROPIC_API_KEY).
+        if dynamic_api_key is not None:
+            agentic_loop_params["api_key"] = dynamic_api_key
+        if dynamic_api_base is not None:
+            agentic_loop_params["api_base"] = dynamic_api_base
+        litellm_logging_obj.model_call_details["agentic_loop_params"] = agentic_loop_params
 
         # Check if stream was converted for WebSearch interception
         # This is set in the async wrapper above when stream=True is converted to stream=False

litellm/llms/custom_httpx/llm_http_handler.py

@@ -4642,11 +4642,15 @@ class BaseLLMHTTPHandler:
             raise ValueError("Agentic loop plan missing patched messages")
 
         full_model_name = model
+        agentic_api_key: Optional[str] = None
+        agentic_api_base: Optional[str] = None
         if logging_obj is not None:
             agentic_params = logging_obj.model_call_details.get(
                 "agentic_loop_params", {}
             )
             full_model_name = cast(str, agentic_params.get("model", model))
+            agentic_api_key = agentic_params.get("api_key")
+            agentic_api_base = agentic_params.get("api_base")
 
         optional_params = dict(anthropic_messages_optional_request_params)
         optional_params.update(patch.optional_params)
@@ -4681,6 +4685,8 @@ class BaseLLMHTTPHandler:
                 "messages": patch.messages,
                 "model": patch.model or full_model_name,
                 "stream": stream,
+                **({"api_key": agentic_api_key} if agentic_api_key else {}),
+                **({"api_base": agentic_api_base} if agentic_api_base else {}),
                 **optional_params,
                 **kwargs_for_followup,
             }