Cognitive Test Orchestration in Cloud Environments using MCP–LLM Hybrid Agents

Abstract

The increasing complexity of distributed applications has outpaced traditional test automation strategies, creating an urgent need for intelligent, self-adaptive approaches that can ensure reliability in dynamic cloud environments. This paper introduces a novel MCP–LLM Hybrid Orchestration Framework, which integrates the Model Control Protocol (MCP) with Large Language Model (LLM)-driven agents to enable cognitive test scheduling, contextual learning, and self-healing automation within Continuous Integration/Continuous Deployment (CI/CD) pipelines. The proposed design empowers automation systems to make autonomous decisions based on environmental signals, telemetry feedback, and real-time system resource states.

Unlike traditional orchestration frameworks that rely on static rules, the hybrid MCP-LLM architecture leverages semantic reasoning and data-driven insights to dynamically optimize test selection and prioritization. Experimental evaluations conducted on Google Cloud Platform (GCP) demonstrate a 34% improvement in test execution efficiency and a 27% reduction in cloud resource utilization when compared to conventional Playwright-based automation. The results confirm that this approach not only enhances performance but also establishes a foundation for context-aware, AI-augmented quality engineering. Ultimately, this framework represents a key step toward autonomous, cognitive software testing ecosystems capable of evolving alongside modern distributed architectures.