diff --git a/letta/llm_api/anthropic_client.py b/letta/llm_api/anthropic_client.py
index 8acd65dd..7f133afb 100644
--- a/letta/llm_api/anthropic_client.py
+++ b/letta/llm_api/anthropic_client.py
@@ -356,9 +356,17 @@ class AnthropicClient(LLMClientBase):
     ) -> Union[anthropic.AsyncAnthropic, anthropic.Anthropic]:
         api_key, _, _ = self.get_byok_overrides(llm_config)
 
+        # For MiniMax provider, use minimax_api_key from settings
+        if not api_key and llm_config.provider_name == "minimax":
+            api_key = model_settings.minimax_api_key
+
         # For claude-pro-max provider, use OAuth Bearer token instead of api_key
         is_oauth_provider = llm_config.provider_name == "claude-pro-max"
 
+        # Only use custom base_url for MiniMax (Anthropic-compatible API)
+        # The Anthropic SDK adds /v1/messages internally, so we only override for non-Anthropic providers
+        base_url = llm_config.model_endpoint if llm_config.provider_name == "minimax" else None
+
         if async_client:
             if api_key:
                 if is_oauth_provider:
@@ -370,8 +378,8 @@ class AnthropicClient(LLMClientBase):
                             "anthropic-beta": "oauth-2025-04-20",
                         },
                     )
-                return anthropic.AsyncAnthropic(api_key=api_key, max_retries=model_settings.anthropic_max_retries)
-            return anthropic.AsyncAnthropic(max_retries=model_settings.anthropic_max_retries)
+                return anthropic.AsyncAnthropic(api_key=api_key, base_url=base_url, max_retries=model_settings.anthropic_max_retries)
+            return anthropic.AsyncAnthropic(base_url=base_url, max_retries=model_settings.anthropic_max_retries)
 
         if api_key:
             if is_oauth_provider:
@@ -383,8 +391,8 @@ class AnthropicClient(LLMClientBase):
                         "anthropic-beta": "oauth-2025-04-20",
                     },
                 )
-            return anthropic.Anthropic(api_key=api_key, max_retries=model_settings.anthropic_max_retries)
-        return anthropic.Anthropic(max_retries=model_settings.anthropic_max_retries)
+            return anthropic.Anthropic(api_key=api_key, base_url=base_url, max_retries=model_settings.anthropic_max_retries)
+        return anthropic.Anthropic(base_url=base_url, max_retries=model_settings.anthropic_max_retries)
 
     @trace_method
     async def _get_anthropic_client_async(
@@ -392,9 +400,17 @@ class AnthropicClient(LLMClientBase):
     ) -> Union[anthropic.AsyncAnthropic, anthropic.Anthropic]:
         api_key, _, _ = await self.get_byok_overrides_async(llm_config)
 
+        # For MiniMax provider, use minimax_api_key from settings
+        if not api_key and llm_config.provider_name == "minimax":
+            api_key = model_settings.minimax_api_key
+
         # For claude-pro-max provider, use OAuth Bearer token instead of api_key
         is_oauth_provider = llm_config.provider_name == "claude-pro-max"
 
+        # Only use custom base_url for MiniMax (Anthropic-compatible API)
+        # The Anthropic SDK adds /v1/messages internally, so we only override for non-Anthropic providers
+        base_url = llm_config.model_endpoint if llm_config.provider_name == "minimax" else None
+
         if async_client:
             if api_key:
                 if is_oauth_provider:
@@ -406,8 +422,8 @@ class AnthropicClient(LLMClientBase):
                             "anthropic-beta": "oauth-2025-04-20",
                         },
                     )
-                return anthropic.AsyncAnthropic(api_key=api_key, max_retries=model_settings.anthropic_max_retries)
-            return anthropic.AsyncAnthropic(max_retries=model_settings.anthropic_max_retries)
+                return anthropic.AsyncAnthropic(api_key=api_key, base_url=base_url, max_retries=model_settings.anthropic_max_retries)
+            return anthropic.AsyncAnthropic(base_url=base_url, max_retries=model_settings.anthropic_max_retries)
 
         if api_key:
             if is_oauth_provider:
@@ -419,8 +435,8 @@ class AnthropicClient(LLMClientBase):
                         "anthropic-beta": "oauth-2025-04-20",
                     },
                 )
-            return anthropic.Anthropic(api_key=api_key, max_retries=model_settings.anthropic_max_retries)
-        return anthropic.Anthropic(max_retries=model_settings.anthropic_max_retries)
+            return anthropic.Anthropic(api_key=api_key, base_url=base_url, max_retries=model_settings.anthropic_max_retries)
+        return anthropic.Anthropic(base_url=base_url, max_retries=model_settings.anthropic_max_retries)
 
     @trace_method
     def build_request_data(
diff --git a/letta/schemas/providers/base.py b/letta/schemas/providers/base.py
index 6b5b722d..f24794d6 100644
--- a/letta/schemas/providers/base.py
+++ b/letta/schemas/providers/base.py
@@ -192,6 +192,7 @@ class Provider(ProviderBase):
             GroqProvider,
             LettaProvider,
             LMStudioOpenAIProvider,
+            MiniMaxProvider,
             MistralProvider,
             OllamaProvider,
             OpenAIProvider,
@@ -244,6 +245,8 @@ class Provider(ProviderBase):
                 return LMStudioOpenAIProvider(**self.model_dump(exclude_none=True))
             case ProviderType.bedrock:
                 return BedrockProvider(**self.model_dump(exclude_none=True))
+            case ProviderType.minimax:
+                return MiniMaxProvider(**self.model_dump(exclude_none=True))
             case _:
                 raise ValueError(f"Unknown provider type: {self.provider_type}")
 
diff --git a/letta/schemas/providers/minimax.py b/letta/schemas/providers/minimax.py
index 616cf11d..70777113 100644
--- a/letta/schemas/providers/minimax.py
+++ b/letta/schemas/providers/minimax.py
@@ -73,7 +73,7 @@ class MiniMaxProvider(Provider):
             configs.append(
                 LLMConfig(
                     model=model["name"],
-                    model_endpoint_type="minimax",
+                    model_endpoint_type="anthropic",
                     model_endpoint=self.base_url,
                     context_window=model["context_window"],
                     handle=self.get_handle(model["name"]),
diff --git a/letta/server/server.py b/letta/server/server.py
index bfdb21a8..7cb815a8 100644
--- a/letta/server/server.py
+++ b/letta/server/server.py
@@ -67,6 +67,7 @@ from letta.schemas.providers import (
     GroqProvider,
     LettaProvider,
     LMStudioOpenAIProvider,
+    MiniMaxProvider,
     OllamaProvider,
     OpenAIProvider,
     OpenRouterProvider,
@@ -342,6 +343,13 @@ class SyncServer(object):
                     api_key_enc=Secret.from_plaintext(model_settings.xai_api_key),
                 )
             )
+        if model_settings.minimax_api_key:
+            self._enabled_providers.append(
+                MiniMaxProvider(
+                    name="minimax",
+                    api_key_enc=Secret.from_plaintext(model_settings.minimax_api_key),
+                )
+            )
         if model_settings.zai_api_key:
             self._enabled_providers.append(
                 ZAIProvider(
diff --git a/tests/test_providers.py b/tests/test_providers.py
index 21600682..3b013a35 100644
--- a/tests/test_providers.py
+++ b/tests/test_providers.py
@@ -154,8 +154,8 @@ async def test_minimax():
         assert model.context_window == 200000
         # All MiniMax models have 128K max output
         assert model.max_tokens == 128000
-        # All use minimax endpoint type
-        assert model.model_endpoint_type == "minimax"
+        # MiniMax uses Anthropic-compatible API endpoint
+        assert model.model_endpoint_type == "anthropic"
 
 
 @pytest.mark.skipif(model_settings.azure_api_key is None, reason="Only run if AZURE_API_KEY is set.")