AI BriefingStream Vessel implements per-user MCP redirect allowlists to prevent OAuth phishing
·2026.11.04 09:00
Key point
The design validates redirect URIs at the authorization step rather than registration, allowing users to trust specific https addresses while blocking open-redirect attacks.
Details
- Stream Vessel introduces a per-user trust model for MCP OAuth clients, allowing users to approve up to 10 specific https redirect addresses for their own accounts without requiring operator-level configuration changes.
- A May 2026 study found that 95.8% of 119 tested OAuth-enabled MCP servers accepted arbitrary redirect URIs from anonymous requesters, creating a high-risk vector for one-click account takeover via phishing.
- The system validates redirect URIs at the authorization page where the user is authenticated, rather than at registration, ensuring that authorization codes and error messages are never sent to untrusted addresses.
- Security controls include requiring users to type the host to trust it, emailing notifications for every addition, and immediately revoking grants and in-flight codes when an address is removed.
- The approach addresses the limitations of fixed operator allowlists, which block legitimate custom agents, and dynamic client registration, which proves publisher identity but not user intent.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Carrot Launches Korea's First Public Beta for Advertising MCP, Sharing a Case Study on Expanding Developer Roles
Anthropic amended S-1 reveals $17 billion Q3 revenue and $895.4 billion in compute commitmentsPierre Computer Company Teases Upcoming Announcement