requestClientSampling method

Future<Map<String, dynamic>> requestClientSampling(String sessionId, Map<String, dynamic> params, { Duration timeout = const Duration(seconds: 60), })

Server-initiated request: ask the connected client's LLM to generate a completion (spec sampling/createMessage).

params follows the spec CreateMessageRequest.params shape: messages, maxTokens (required), plus optional modelPreferences, systemPrompt, includeContext, temperature, stopSequences, metadata.

Returns the spec CreateMessageResult map (role, content, model, optional stopReason).

The client must advertise the sampling capability during initialize; otherwise this throws McpError with methodNotFound.

Implementation

Future<Map<String, dynamic>> requestClientSampling(
  String sessionId,
  Map<String, dynamic> params, {
  Duration timeout = const Duration(seconds: 60),
}) async {
  final session = _sessions[sessionId];
  if (session == null) {
    throw StateError('Unknown sessionId: $sessionId');
  }
  final clientHasSampling = session.capabilities?['sampling'] != null;
  if (!clientHasSampling) {
    throw McpError(
      'Client does not advertise the `sampling` capability',
      code: ErrorCode.methodNotFound,
    );
  }
  final result = await _sendRequestToClient(
      sessionId, 'sampling/createMessage', params, timeout: timeout);
  return result is Map<String, dynamic>
      ? result
      : Map<String, dynamic>.from(result as Map);
}