he Model Context Protocol (MCP), introduced by Anthropic in 2024, has become the dominant interface for connecting AI agents to external tools and data sources, with ecosystem-wide adoption exceeding 57 million weekly package downloads. Despite this rapid adoption, MCP lacks native cryptographic security at the message level. This paper presents MCPS (MCP Secure), a backward-compatible cryptographic signing layer that provides per-message ECDSA P-256 signatures, replay protection, tool integrity verification, and a portable agent identity mechanism. We describe the protocol design, analyse the threat model, and demonstrate sub-2ms overhead per message. An open-source reference implementation is available as a zero-dependency npm package.