Model Context Protocol Server for Hybrid Quantum-High Performance Computing Environments

This paper proposes an AI-driven framework based on the Model Context Protocol (MCP), enabling large language model agents to autonomously execute quantum computing workflows. Through the OpenQASM interpretation pipeline, CUDA-Q automation workflow, and remote quantum hardware asynchronous execution pipeline, it achieves the automatic execution of quantum algorithm primitives, bridging the gap between AI execution capabilities and the complex environment of quantum computing.

Large language models are transforming the scientific research paradigm, and the concept of "AI scientists" is gradually becoming a reality. However, there is a key execution gap: AI cannot actually perform experiments that require specialized hardware and complex resource management, especially in the quantum computing field—such as generating correct code, managing QPUs, and coordinating HPC clusters—which has become a bottleneck restricting the development of "AI quantum scientists".

The research team proposes a framework based on the Model Context Protocol (MCP) and develops a dedicated MCP server as an intermediary between AI and the quantum-classical hybrid environment. The core components include: 1. OpenQASM code interpretation pipeline (parses and validates code, returns error feedback); 2. CUDA-Q automation workflow (deploys to the ABCI-Q platform, coordinates resource allocation); 3. Remote quantum hardware asynchronous execution pipeline (optimized for the Quantinuum simulator, supports asynchronous task submission and result callbacks).

The framework's capabilities are verified through two demonstrations: 1. Quantum sampling: The AI describes requirements in natural language (e.g., 1000 measurements of an 8-qubit GHZ state), and the system automatically generates OpenQASM code, executes it, and returns results; 2. Expectation value calculation: The AI requests the calculation of the expectation value of an observable (e.g., Hamiltonian H=Z0*Z1+X2), and the system constructs the circuit, performs measurements, processes statistical errors, and supports variational quantum algorithm optimization.

The technical contributions include: at the protocol level (among the first to apply MCP to quantum computing), at the engineering level (solving practical automation challenges), and at the application level (enabling end-to-end AI control of hybrid environments). The significance for the quantum computing community: providing a rapid prototyping platform for algorithm researchers, a standardized AI interface for hardware developers, and lowering the threshold for quantum programming for educators.

Current limitations: mainly supports quantum simulators, real hardware support needs improvement; error handling mechanisms need enhancement. Future directions: expand quantum programming languages and platforms, enhance AI's ability to autonomously design and optimize algorithms, explore multi-agent collaboration models. From a broader perspective, this work represents the trend of integration between AI and scientific infrastructure, which can be extended to other scientific research fields to accelerate AI penetration.