MODEL CONTEXT PROTOCOL (MCP) AND SUPPLY-CHAIN SECURITY

MONITORING THAT LINKS PROMPTS TO TOOL CALLS AND ACTIONS Prevention will not catch everything, so monitoring is the safety net. The idea is to record a connected story: which prompt or document led to which tool call, and which downstream action followed. This uses the tracing ideas from Chapter 28. With such links, a security team can notice that a certain kind of message is regularly followed by a sensitive tool call, which may mean an attacker is refining an injection. A SAFE ROUTINE FOR ADDING A NEW SERVER Adding a new Model Context Protocol server to your harness is like hiring a contractor who will work inside your building. Follow a short routine each time. First, identify the publisher and confirm that it is who it claims to be, using the official page and not a link from a message. Second, read the list of tools the server offers and the permissions it asks for, and refuse anything broader than the job needs. Third, pin the exact version so that an update cannot change behavior without your knowledge. Fourth, scan the code and the packages it depends on for known problems. Fifth, run it first in a sandbox with no access to real data and watch what it does. Sixth, add it to your inventory with the name of a person who is responsible for it. Finally, schedule a regular review, because a server that was safe last year may have changed owners or code since.