An easy, realistic model for MCP connectivity

Lee Briggs, Director of Solutions Engineering at Tailscale, discusses the evolution and security challenges of the Model Context Protocol (MCP) in the context of integrating local information into Large Language Models (LLMs). Briggs highlights the rapid development of MCP, noting that initial designs focused on secure local connectivity through stdio transport, with later iterations introducing remote access via Server Side Events (SSE) and eventually Streamable HTTP. Despite advancements, Briggs expresses concerns about security vulnerabilities, particularly the lack of robust authentication measures and potential exposure of sensitive data on the internet. Drawing from his experience at Tailscale, he explores using Tailscale's application awareness to enhance MCP's security model, proposing a setup that leverages Tailscale for secure, private connections. Briggs acknowledges the tension between participating in technological advancements and addressing privacy concerns, emphasizing his commitment to building more secure systems while questioning the broader implications of data accessibility to LLMs.