Quantifying Conversation Drift in MCP via Latent Polytope

Haoran Shi, Hongwei Yao, Shuo Shao +4

The Model Context Protocol (MCP) enhances large language models (LLMs) by integrating external tools, enabling dynamic aggregation of real-time data to improve task execution. Howe…