INVALID_ARGS

INVALID_ARGS

Error message

Cannot use multiple custom provider flags simultaneously. Choose only one: openrouter, ollama, bedrock, azure, vertex, or openai-compatible.

What it means

The models tool accepts at most one custom provider flag (openrouter, ollama, bedrock, azure, vertex, or openai-compatible). When two or more are supplied simultaneously, modelsDirect returns INVALID_ARGS, since only a single provider context can be queried per call.

Source

Thrown at mcp-server/src/core/direct-functions/models.js:82

	// Create a logger wrapper that the core functions can use
	const mcpLog = createLogWrapper(log);

	log.info(`Executing models_direct with args: ${JSON.stringify(args)}`);
	log.info(`Using project root: ${projectRoot}`);

	// Validate flags: only one custom provider flag can be used simultaneously
	const customProviderFlags = CUSTOM_PROVIDERS_ARRAY.filter(
		(provider) => args[provider]
	);

	if (customProviderFlags.length > 1) {
		log.error(
			'Error: Cannot use multiple custom provider flags simultaneously.'
		);
		return {
			success: false,
			error: {
				code: 'INVALID_ARGS',
				message:
					'Cannot use multiple custom provider flags simultaneously. Choose only one: openrouter, ollama, bedrock, azure, vertex, or openai-compatible.'
			}
		};
	}

	try {
		enableSilentMode();

		try {
			// Check for the listAvailableModels flag
			if (args.listAvailableModels === true) {
				return await getAvailableModelsList({
					session,
					mcpLog,
					projectRoot
				});
			}

View on GitHub (pinned to c0c98d367c)

Solutions

Send exactly one provider flag per call; issue separate calls for additional providers
Call models with no flags to list all default models, then filter client-side
Fix prompting/tool-schema guidance so the model selects only one flag

Example fix

// before
await mcp.call('models', { openrouter: true, ollama: true });
// after
await mcp.call('models', { openrouter: true });
await mcp.call('models', { ollama: true }); // separate call

Defensive patterns

Strategy: validation

Validate before calling

const PROVIDER_FLAGS = ['openrouter','ollama','bedrock','azure','vertex','openai-compatible'];
function assertSingleProvider(args) {
  const selected = PROVIDER_FLAGS.filter((f) => args[f] === true);
  if (selected.length > 1) {
    throw new Error(`Only one provider flag allowed; got: ${selected.join(', ')}`);
  }
  return args;
}

Type guard

function hasSingleProviderFlag(args) {
  const flags = ['openrouter','ollama','bedrock','azure','vertex','openai-compatible'];
  return flags.filter((f) => args?.[f] === true).length <= 1;
}

Try / catch

const res = await callTool('models', args);
if (!res.success && res.error?.code === 'INVALID_ARGS') {
  console.error('Provider flag conflict:', res.error.message);
  // retry with exactly one flag
}
// or: catch (e) { if (e.message.includes('provider flags')) { ... } }

Prevention

Model provider choice as a single enum, not independent booleans
When LLMs compose tool calls, instruct single-select for provider flags
Issue one call per provider instead of combining flags
Validate flag counts client-side before dispatching the models call

When it happens

Trigger: Calling the models tool with e.g. both openrouter:true and ollama:true (any combination of two or more provider flags).

Common situations: LLM-generated tool calls combining flags from a list of options; UI toggles that don't enforce single-select; copying a previous call and adding a second provider flag; misunderstanding flags as additive filters.

Related errors

AI-assisted analysis of eyaltoledano/claude-task-master@c0c98d367c (2026-08-29). Data as JSON: /api/errors/168a51e7253e4def. Report an issue: GitHub.