Understanding How Enterprises Adopt the Model Context Protocol for LLM-Driven Software Engineering.
Abstract
Large Language Models (LLMs) are increasingly used in AI-based software engineering, but their limitations in complex task execution and multi-tool coordination have driven growing interest in the Model Context Protocol (MCP). Existing research has mainly focused on MCP's technical design, with limited empirical evidence on how it is adopted and used in enterprise practice, particularly with regard to deployment challenges, operational risks, and practitioner expectations. To address this gap, we conducted semi-structured interviews with 20 practitioners from eight companies in the Internet and financial sectors. The findings show that MCP is valued for supporting cross-system collaboration, task decoupling, and knowledge reuse in LLM-based workflows, but its adoption remains constrained by ecosystem fragmentation, cross-component coordination difficulties, and unresolved problems in distributed state management and fault diagnosis. Participants also expressed strong demand for better standardization, lower adoption barriers through low-code or pluginbased approaches, and more systematic operational support. These results provide early empirical evidence on enterprise MCP practice and offer practical implications for improving MCP's standardization, usability, and deployment readiness in real-world software engineering environments. Keywords-Model Context Protocol, AI Protocol, Large Language Models, Software Security, Semi-Structured Interviews 1. I NTRODUCTION Large language models (LLMs) have been widely adopted in enterprise settings, but they still face notable limitations in handling complex tasks such as cross-system invocation and multi-tool collaboration [1]. To address these challenges, Anthropic proposed the Model Context Protocol (MCP) in late 2024 [2]. Since then, MCP has emerged as an intermediatelayer protocol for connecting LLMs with external tools and resources [1]. MCP is built around four core components-Host, Client, Server, and Resource-which together support standardized tool invocation and context exchange between LLMs and heterogeneous systems [3]. Current research on MCP has primarily focused on three areas. First, benchmarking studies