- Post History
- Subscribe to RSS Feed
- Mark as New
- Mark as Read
- Bookmark
- Subscribe
- Printer Friendly Page
- Report Inappropriate Content
4 hours ago - edited 3 hours ago
What's new in AI Gateway?
This release adds client access control, giving AI Stewards a way to limit exactly which tools a connected client can reach on an MCP Server, instead of granting every registered client access to everything.
By default, any client registered for an MCP Server can call every active tool it exposes. That's fine for a lot of use cases, but not every client needs, or should have, the same level of access. A reporting tool that only needs to read data shouldn't also be able to trigger a write action just because it's connected to the same server.
This release adds a fifth capability to AI Gateway's Control pillar, alongside authentication, pause controls, tool scanning, and the Data Sensitivity Check.
For more configuration information, see the product documentation.
Jump to
New this release
Client access control lets AI Stewards restrict which MCP Server tools a given client can reach. Instead of every connected client having the same blanket access, each client only gets the specific tools it's been granted, nothing more.
A client access policy assigns one client to one or more tools on a server. A client with no policy on a server has no access to its tools at all once client access control is turned on there, there's no default fallback to "everything."
How it's enforced
AI Gateway checks each client's access token against its policies and only allows the tools those policies grant. A tool that's inactive stays blocked even if a policy includes it.
Worth knowing
Policy changes can take up to 30 minutes to reach clients that are already connected, since an active client keeps its current access until its token renews.
|
BEFORE Any client registered for an MCP Server could call every active tool on it, there was no way to scope access by client. |
NOW AI Stewards grant each client only the tools it needs, a client with no policy has no tool access on that server at all. |
Summary: Tool access is now scoped per client instead of applying equally to every client connected to a server.
The Access management card on an MCP Server's AI Gateway > Policies tab
Feature summary
| Feature | Role(s) | Available |
|---|---|---|
| MCP Server and Tool Catalog | AI Steward, Product Owner, Developer | September 2026 |
| Lifecycle Management and Approvals | AI Steward, Product Owner, Compliance & Security | September 2026 |
| Authentication and Connectivity | AI Steward, Developer, Compliance & Security | September 2026 |
| Runtime Policy Enforcement (Pause) | AI Steward, Compliance & Security | September 2026 |
| Tool Scanning and the Tool Catalog | AI Steward, Compliance & Security | September 2026 |
| Data Sensitivity Check | AI Steward, Compliance & Security | September 2026 |
| Operational Visibility | AI Steward, Product Owner, Compliance & Security | September 2026 |
| Client access control | AI Steward, Platform Admin, CISO | November 2026 |
For more on AI Gateway, including how client access control works and common questions, see the AI Gateway FAQ →