From 083990f6d7065e456f450901e02d3aa251c3218d Mon Sep 17 00:00:00 2001
From: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
Date: Mon, 2 Nov 2026 13:05:16 -0800
Subject: [PATCH] fix(anthropic): run the advisor loop for an anthropic
 deployment on a Foundry host

The advisor interceptor treated every anthropic deployment as native, so a Foundry-hosted one
received the advisor_20260301 tool and Foundry answered 400. Interceptors now get the
deployment's beta row from resolve_beta_provider, so that deployment runs litellm's advisor
loop the same way an azure_ai deployment does
---
 .../llms/anthropic/messages/native/handler.py |  9 ++++-
 .../messages/native/interceptors/README.md    |  4 +-
 .../native/test_advisor_integration.py        | 40 +++++++++++++++++++
 3 files changed, 49 insertions(+), 4 deletions(-)

diff --git a/litellm/llms/anthropic/messages/native/handler.py b/litellm/llms/anthropic/messages/native/handler.py
index 922de386fe4..37417806b54 100644
--- a/litellm/llms/anthropic/messages/native/handler.py
+++ b/litellm/llms/anthropic/messages/native/handler.py
@@ -15,7 +15,7 @@ from typing import Final, cast
 from pydantic import TypeAdapter
 
 import litellm
-from litellm.anthropic_beta_headers_manager import aload_beta_headers_config
+from litellm.anthropic_beta_headers_manager import aload_beta_headers_config, resolve_beta_provider
 from litellm.constants import PROVIDERS_THAT_AUTHENTICATE_ON_PROVIDER_INFO
 from litellm.integrations.anthropic_cache_control_hook import AnthropicCacheControlHook
 from litellm.litellm_core_utils.core_helpers import as_str_mapping
@@ -540,11 +540,16 @@ async def anthropic_messages(
     if short_circuit_response is not None:
         return short_circuit_response
 
+    serving_provider: Final = (
+        resolve_beta_provider(custom_llm_provider, AnthropicModelInfo.get_api_base(api_base))
+        if custom_llm_provider
+        else None
+    )
     # Run registered MessagesInterceptors (e.g. advisor orchestration loop).
     # Named params on `anthropic_messages` are bound to locals, not `**kwargs`,
     # so forward them explicitly — otherwise interceptor sub-calls drop them.
     for interceptor in get_messages_interceptors():
-        if interceptor.can_handle(tools, custom_llm_provider):
+        if interceptor.can_handle(tools, serving_provider):
             return await interceptor.handle(
                 model=model,
                 messages=messages,
diff --git a/litellm/llms/anthropic/messages/native/interceptors/README.md b/litellm/llms/anthropic/messages/native/interceptors/README.md
index b6df1edc854..f62e03ed43d 100644
--- a/litellm/llms/anthropic/messages/native/interceptors/README.md
+++ b/litellm/llms/anthropic/messages/native/interceptors/README.md
@@ -51,9 +51,9 @@ _interceptors = [
 
 ### `AdvisorOrchestrationHandler`
 
-Handles `advisor_20260301` tool for providers that don't support it natively (all non-Anthropic providers for now).
+Handles `advisor_20260301` tool for providers that don't support it natively (every provider but Anthropic itself, including an `anthropic/` deployment on an Azure AI Foundry host).
 
-**Triggers when:** `advisor_20260301` is in `tools` AND `custom_llm_provider` is not in `ADVISOR_NATIVE_PROVIDERS`.
+**Triggers when:** `advisor_20260301` is in `tools` AND the deployment's beta row (`resolve_beta_provider`, which maps an `anthropic` deployment on a Foundry host to `azure_ai`) is not in `ADVISOR_NATIVE_PROVIDERS`.
 
 **What it does:**
 - Translates the advisor tool to a regular function tool the provider understands.
diff --git a/tests/unit/llms/anthropic/messages/native/test_advisor_integration.py b/tests/unit/llms/anthropic/messages/native/test_advisor_integration.py
index 8bd36c6564a..b3b8624d472 100644
--- a/tests/unit/llms/anthropic/messages/native/test_advisor_integration.py
+++ b/tests/unit/llms/anthropic/messages/native/test_advisor_integration.py
@@ -190,6 +190,46 @@ async def test_anthropic_provider_bypasses_interceptor():
     assert any("Native anthropic" in b.get("text", "") for b in text_blocks)
 
 
+@pytest.mark.asyncio
+@pytest.mark.parametrize(
+    "api_base",
+    ["https://res.services.ai.azure.com/anthropic", "https://res.cognitiveservices.azure.com/anthropic"],
+)
+async def test_anthropic_deployment_on_foundry_runs_the_advisor_loop(api_base):
+    from litellm.llms.anthropic.messages.native.handler import (
+        anthropic_messages,
+    )
+
+    sent_tool_types = []
+
+    async def mock_handler(model, messages, tools, stream, max_tokens, **kwargs):
+        sent_tool_types.append([t.get("type") for t in tools or []])
+        if len(sent_tool_types) == 1:
+            return _advisor_call_resp()
+        if len(sent_tool_types) == 2:
+            return _text_resp("Use trial division.", model="claude-opus-4-6")
+        return _text_resp("def is_prime(n): ...")
+
+    with patch(
+        "litellm.llms.anthropic.messages.native.interceptors.advisor._call_messages_handler",
+        side_effect=mock_handler,
+    ):
+        result = await anthropic_messages(
+            model="claude-sonnet-4-6",
+            messages=MESSAGES,
+            tools=[ADVISOR_TOOL],
+            stream=False,
+            max_tokens=512,
+            custom_llm_provider="anthropic",
+            api_base=api_base,
+        )
+
+    assert len(sent_tool_types) == 3
+    assert all("advisor_20260301" not in types for types in sent_tool_types)
+    assert isinstance(result, dict)
+    assert not [b for b in result.get("content", []) if b.get("type") == "tool_use" and b.get("name") == "advisor"]
+
+
 # ---------------------------------------------------------------------------
 # 4. Regression: top-level named params must be forwarded into executor sub-call
 # ---------------------------------------------------------------------------