How can the agent effectively debug a problem?

[Ghostbird]

Currently I'm stuck on this issue. I'd like my agent to debug an issue I'm having, I've told the agent the steps to reproduce it, but it seems the agent doesn't have any access to the debug console, nor does this MCP seem to provide such access. As such, the agent is flying blind. It can place breakpoints, evaluate expressions, etc, but since it cannot see the errors and output of the thing it's debugging, it cannot meaningfully decide which steps might be sensible.

As an analogy, where I'm the agent, it feels like this MCP gave me a box with tools, and I need fix that the floor in the kitchen is wet, but it doesn't actually permit me to enter the house. But it will tell me exactly what a highly specific spot in the house looks like when asked. It might be that there's actually a tap in the bathroom upstairs that's left open, but there's no real way to figure out that that's the problem in the first place.

Either I need another MCP to complement this one, or I'm seriously failing to understand something. Currently I'm using Claude Sonnet, but it completely fails to use this MCP to do anything meaningful, beyond starting the debug session.

[Ghostbird]

I've found a workaround specifically for web debugging. If you enable workbench.browser.enableChatTools and launch into the integrated browser, the agent can access the browser console using playwright.