Your site says an agent can use it. This runs a real task through your WebMCP tools and checks the environment afterwards to see whether anything actually happened.
Pick a target and a task, then run it. The Lab reads the tools the page actually publishes, checks your task against them, executes it, and then asks the site itself what changed.
This is the real page, running its own WebMCP tools. Watch it change as the evaluation executes.