Toward Context-Coherent Agentic Computing: MCP Servers as a Semantic Consistency Layer for Tool-Using Autonomous Systems

This paper proposes a context-coherent architecture for agentic computing, introducing MCP servers as a semantic consistency layer for tool-using autonomous systems. We analyze context drift, reliability risks, prompt injection threats, and governance challenges, and present a formal model and implementation blueprint for maintaining consistent, auditable agent execution across distributed tools. The work explores consistency levels, provenance tracking, security enforcement, and runtime validation mechanisms necessary for reliable large-scale agent deployments. This research contributes to emerging discussions on agent infrastructure and operational safety in AI systems.