What causes "%s is not a valid type of embedding model" in googleapis/mcp-toolbox?

Set `type: gemini` and put the actual model name (e.g. text-embedding-004) in the `model` field

%s is not a valid type of embedding model

Error message

%s is not a valid type of embedding model

What it means

The embedding model entry has a string `type`, but its value is not the only supported embedding model type, gemini.EmbeddingModelType ("gemini"). UnmarshalYAMLEmbeddingModelConfig only accepts gemini and rejects all other type strings.

Source

Thrown at internal/server/config.go:443

				return nil, fmt.Errorf("`introspectionParamName` is not allowed when `mcpEnabled` is false")
			}
			if len(actual.ScopesRequired) > 0 {
				return nil, fmt.Errorf("`scopesRequired` is not allowed when `mcpEnabled` is false")
			}
		}
		return actual, nil
	default:
		return nil, fmt.Errorf("%s is not a valid type of auth service", resourceType)
	}
}

func UnmarshalYAMLEmbeddingModelConfig(ctx context.Context, name string, r map[string]any) (embeddingmodels.EmbeddingModelConfig, error) {
	resourceType, ok := r["type"].(string)
	if !ok {
		return nil, fmt.Errorf("missing 'type' field or it is not a string")
	}
	if resourceType != gemini.EmbeddingModelType {
		return nil, fmt.Errorf("%s is not a valid type of embedding model", resourceType)
	}
	dec, err := util.NewStrictDecoder(r)
	if err != nil {
		return nil, fmt.Errorf("error creating decoder: %s", err)
	}
	actual := gemini.Config{Name: name}
	if err := dec.DecodeContext(ctx, &actual); err != nil {
		return nil, fmt.Errorf("unable to parse as %q: %w", name, err)
	}
	return actual, nil
}

func UnmarshalYAMLToolConfig(ctx context.Context, name string, r map[string]any) (tools.ToolConfig, error) {
	err := NameValidation(name)
	if err != nil {
		return nil, err
	}
	resourceType, ok := r["type"].(string)

View on GitHub (pinned to 8cc6e09de2)

Solutions

Set `type: gemini` and put the actual model name (e.g. text-embedding-004) in the `model` field
Check docs for the currently supported embedding model types before adding new providers

Example fix

// before
embeddingModels:
  my-embed:
    type: openai
    model: text-embedding-3-small
// after
embeddingModels:
  my-embed:
    type: gemini
    model: text-embedding-004

Defensive patterns

Strategy: validation

Validate before calling

func validateEmbeddingModelType(cfg map[string]any) error {
  t, _ := cfg["type"].(string)
  if t != "gemini" { return fmt.Errorf("embedding model type must be 'gemini', got %q", t) }
  return nil
}

Type guard

func isGeminiEmbedding(cfg map[string]any) bool {
  t, ok := cfg["type"].(string)
  return ok && t == "gemini"
}

Prevention

Remember gemini is currently the only embedding model type; put the model name in 'model'
Do not reuse LLM provider configs for embeddingModels
Verify against docs for the toolbox version in use

When it happens

Trigger: UnmarshalPrimitiveConfig encounters an embeddingModels entry with e.g. `type: openai`, `type: vertex`, `type: gemini-embedding`, or any misspelling of `gemini`.

Common situations: Copying an LLM model config (which may allow other providers) into embeddingModels; assuming OpenAI/Vertex embeddings are supported; writing the full model name in `type` instead of using the `model` field.

Related errors

AI-assisted analysis of googleapis/mcp-toolbox@8cc6e09de2 (2026-09-05). Data as JSON: /api/errors/62f3af8228587408. Report an issue: GitHub.