Have Agent run your project’s test command via run_command, read failures, and patch. Keep MCP browser tests for UI flows when Playwright MCP is connected.