Website fingerprinting attacks, which analyse the metadata of encrypted network communication to identify visited websites, have been shown to be effective on privacy-enhancing technologies including virtual private networks (VPNs) and encrypted proxies. Despite this, VPNs are still undefended against these attacks, leaving millions of users vulnerable. Proposed defences against website fingerprinting require cooperation between the client and a remote endpoint to reshape the network traffic, thereby hindering deployment. We observe that the rapid and wide-spread deployment of QUIC and HTTP/3 creates an exciting opportunity to build website-fingerprinting defences directly into client applications, such as browsers, without requiring any changes to web servers, VPNs, or the deployment of new network services. We therefore design and implement the QCSD framework, which leverages QUIC and HTTP/3 to emulate existing website-fingerprinting defences by bidirectionally adding cover traffic and reshaping connections solely from the client. As case studies, we emulate both the FRONT and Tamaraw defences solely from the client and collected several datasets of live-defended traffic on which we evaluated modern machine-learning based attacks. Our results demonstrate the promise of this approach in shaping connections towards client-orchestrated defences, thereby removing a primary barrier to the deployment of website-fingerprinting defences.