Research Briefing: MCP Security

This report examines the current security posture of the Model Context Protocol (MCP) ecosystem, highlighting risks in local and remote servers, client auto-run behavior, supply-chain and registry trust gaps, and vulnerabilities related to OAuth and transport (e.g., DNS rebinding). It provides pragmatic guidance for users and builders—favoring trusted sources, least privilege, auditing, sandboxing, gateways, and allowlisting—while surveying promising developments such as official registries with signing/pinning, tool namespacing, containerized/WASM isolation, and proxy-based governance. The piece underscores that MCP is rapidly evolving and that defenders must harden defaults now to mitigate issues like typosquatting, impersonation, prompt injection, and RCE.