Letting users trust their own MCP clients, without opening a phishing hole
Summary
The article analyzes OAuth MCP redirect URI vulnerabilities that enable phishing via open redirects due to registration accepting arbitrary redirect addresses. It advocates per-user trust decisions at sign-in, preserves operator allowlists, and introduces a design where a user’s own trusted redirect addresses govern token redirection. It cites research and CVEs and provides a practical checklist for securing MCP deployments.