I have a few questions about the practical nature of this
- How does the protocol specifically protect against the theft of authentication tokens stored on the MCP server, especially if the server itself
- What is the mechanism for vetting third-party tools to prevent "tool poisoning" or the use of malicious tools that could lead to data exfiltration or command injection
- Does the protocol enforce a principle of least privilege? How does it prevent a tool from gaining overly broad permissions and being exploited as a "confused deputy" by a malicious promot?