mlflow/mlflow · error · TypeError

Invalid config type {config.model.config}

Error message

Invalid config type {config.model.config}

What it means

The AnthropicProvider constructor requires the endpoint's model config to be an AnthropicConfig instance. If config.model.config is None or a different config type (e.g., an OpenAI or Cohere config object), a TypeError is raised at endpoint/adapter initialization time, before any request is made.

Source

Thrown at mlflow/gateway/providers/anthropic.py:537

        raise NotImplementedError

    @classmethod
    def model_to_embeddings(cls, resp, config):
        raise NotImplementedError


class AnthropicProvider(BaseProvider, AnthropicAdapter):
    DISPLAY_NAME = "Anthropic"
    CONFIG_TYPE = AnthropicConfig

    PASSTHROUGH_PROVIDER_PATHS = {
        PassthroughAction.ANTHROPIC_MESSAGES: "messages",
    }

    def __init__(self, config: EndpointConfig, enable_tracing: bool = False) -> None:
        super().__init__(config, enable_tracing=enable_tracing)
        if config.model.config is None or not isinstance(config.model.config, AnthropicConfig):
            raise TypeError(f"Invalid config type {config.model.config}")
        self.anthropic_config: AnthropicConfig = config.model.config

    @property
    def headers(self) -> dict[str, str]:
        return {
            "x-api-key": self.anthropic_config.anthropic_api_key,
            "anthropic-version": self.anthropic_config.anthropic_version,
        }

    @property
    def base_url(self) -> str:
        return self.anthropic_config.anthropic_api_base

    @property
    def adapter_class(self) -> type[ProviderAdapter]:
        return AnthropicAdapter

    def _get_headers(

View on GitHub (pinned to 6a27f2decc)

Solutions

1. Fix the gateway endpoint config so the model block includes a valid AnthropicConfig (anthropic_api_key and target_uri for Anthropic).
2. Use AnthropicProvider only with endpoints whose provider is 'anthropic'; match provider class to the config type.
3. Validate your config file (mlflow gateway start) and ensure required Anthropic keys are present before server start.
4. If constructing programmatically, pass AnthropicConfig(...) explicitly instead of None or another provider's config.

Example fix

// before
provider = AnthropicProvider(config=openai_endpoint_config)
// after
from mlflow.gateway.config import AnthropicConfig
provider = AnthropicProvider(config=anthropic_endpoint_config)  # config.model.config is AnthropicConfig

from mlflow.gateway.config import AnthropicConfig
if config.model.config is None or not isinstance(config.model.config, AnthropicConfig):
    raise ValueError("Endpoint config must include an AnthropicConfig model config block before creating AnthropicProvider")

def is_anthropic_config(config) -> bool:
    from mlflow.gateway.config import AnthropicConfig
    return config is not None and isinstance(getattr(config.model, "config", None), AnthropicConfig)

try:
    provider = AnthropicProvider(config=endpoint_config)
except TypeError as e:
    if "Invalid config type" in str(e):
        raise ValueError(
            f"Endpoint '{endpoint_config.name}' is missing an Anthropic config block; "
            "add anthropic_api_key and anthropic provider settings to your gateway config."
        ) from e
    raise

When it happens

Trigger: Instantiating AnthropicProvider (directly or via the gateway server startup/route wiring) with an EndpointConfig whose model.config is None or of the wrong class.

Common situations: A gateway config YAML missing the provider's config block so model.config is None; copying an endpoint config from another provider; programmatic config construction passing the wrong config class; malformed config file not parsed into AnthropicConfig.

Related errors

• trackingUri must be a string
• experimentId must be a string
• Unexpected config type {config.model.config}
• Cannot set both 'temperature' and 'top_p' parameters.
• Invalid value for max_tokens: cannot exceed {MLFLOW_AI_GATEW

AI-assisted analysis of mlflow/mlflow@6a27f2decc (2026-08-29). Data as JSON: /api/errors/30b6e577187dda02. Report an issue: GitHub.