Chrome Releases WebMCP Eval CLI to Test Unpredictable Agent Behavior Before Production

Chrome for Developers has unveiled an experimental WebMCP Eval CLI, a new tool designed to measure and debug how LLM agents interact with website tools—addressing a fundamental gap in agent-centric software testing. Unlike traditional unit tests that verify deterministic code behavior, AI agents produce variable outputs for identical inputs, making it difficult to ensure tools work reliably when called by agents in production.