Model Context Protocol (MCP) is a data transmission protocol developed by Anthropic which can be used to efficiently find and inject context into an LLM’s context window.
MCP facilitates the dynamic definition of tools
Security Concerns
There are a lot of security concerns about MCP in its current early implementation. There are many ways that it could be exploited or used to exfiltrate personal data.
Rug Pull Attacks
The user could register a new MCP tool and then a few days later the tool could update with malicious code that steals their credentials.