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 advice, leak something from its prompt, or produce a convincing phishing draft — but it couldn't t...