Your Agent Has Tools Now: Why MCP Tool Calls Need Runtime Verification

For most of the short history of LLM applications, model risk was text risk. A model could output bad...