← All updates

Security / News explained

An MCP registry upgrade still needs an upstream-address review

The registry's request-forgery fix begins at version 1.35.0. Existing server addresses remain stored, so the November disclosure also calls for configuration and activity review.

By Zyel · Published · 4 min read

AWS's November 11, 2026 bulletin for CVE-2026-125926 identifies a server-side request forgery flaw in mcp-gateway-registry. Versions 1.0.3 through 1.34.0, inclusive, are affected; version 1.35.0 contains the fix. Upgrading is necessary, and stored server addresses still need attention afterward. [1]

Our interpretation is that registering a tool service also makes a network-security decision. The chosen address tells a gateway where to send requests using the gateway's position in the deployment. A registration form may look like directory administration, but its effects travel well beyond the directory.

The gateway carries its own network reach

The maintainers say an authenticated user with permission to register or modify a server could supply an upstream address leading to an internal service. The gateway could then contact that service and sometimes return its response. Depending on network configuration, reachable destinations could include instance metadata and the credentials it provides. [2]

In a hypothetical deployment, a staff member can register a legitimate reporting tool but cannot directly reach a separate internal service. If the gateway can reach both, accepting an arbitrary replacement address could let the staff member borrow that network position. This example explains the boundary; it is not an account of an observed incident.

OWASP describes request forgery as abuse of an application's ability to contact another destination. Its guidance distinguishes applications that need a known set of trusted services from those accepting arbitrary external destinations. It recommends defenses at both the application and network layers, rather than relying on the appearance of a submitted address. [4]

The two registry fixes have different finish lines

The public 1.35.0 release is dated October 29, 2026; the disclosure followed on November 11. That chronology matters when reading an update notice: this is newly published vulnerability detail about an already available release. The release notes also contain deployment-specific upgrade changes, so the security version number should be read alongside the relevant installation notes. [3]

AWS explicitly distinguishes this problem from the registry's separate management-API authorization issue, which was fixed in 1.25.0. Having that earlier fix does not resolve this request-forgery flaw. A deployment in the later affected range still needs the 1.35.0 fix or a subsequent release incorporating it. [1]

The new address validation runs when a server is registered or modified. The maintainers say configurations saved before the upgrade are left unchanged. Existing entries therefore remain part of the response: installing a version with a stronger admission rule does not establish that every older admitted address was intended. [2]

That difference is easy to miss during a busy release. The service starts, familiar tools appear, and ordinary requests succeed. Those observations tell a team something useful about availability. They do not explain who supplied each stored destination or whether it belongs in the gateway's permitted network relationships.

Saved destinations and historical activity need context

AWS calls for examining registered upstream addresses and correcting or removing unintended ones. It also recommends reviewing application logs for unexpected registrations and changes over the period an affected version was deployed. The relevant period follows the installation history, not merely the date an operator first read the bulletin. [1]

Where the registry has an attached cloud role and the log review cannot exclude unexpected activity, the maintainers recommend reviewing that role's activity and rotating its credentials. This is conditional follow-up tied to the deployment and available evidence, rather than proof that every affected installation lost cloud credentials. [2]

Our view is that uncertainty should be recorded precisely. A missing log interval is a gap in what can be ruled out; an unfamiliar server address needs an owner and explanation. Neither observation should be quietly converted into a confirmed breach, and neither should disappear behind a successful software upgrade.

Constrain the network relationship as well as the account

For deployments that cannot immediately upgrade, AWS recommends limiting registration and modification rights to trusted accounts and restricting outbound access from both registry and auth-server components. On EC2, it also recommends requiring IMDSv2 and giving attached roles only the permissions their components need. [1]

EC2 documentation explains that IMDSv2 uses session-oriented requests. An instance can be configured to accept only that version, causing IMDSv1 calls to fail. The operational distinction is between supporting IMDSv2 and requiring it. This is an additional boundary around metadata access, alongside the registry patch and destination controls. [5]

For a gateway that legitimately connects to internal tools, the design question is which relationships are required. OWASP's trusted-destination approach helps frame that decision. Our recommendation is to make those intended connections explicit enough that operators can distinguish a business integration from an unexplained addition without guessing from its hostname. [4]

An MCP gateway can simplify discovery while concentrating useful network access in one place. That makes saved configuration worth the same care as executable code. The patch changes what the registry will accept next; a complete response also accounts for the destinations it already remembers and the activity they may have enabled.

Sources + further reading

Source dates describe the original material. Publication above is the date this Zyel article went live.

  1. 2026-180-AWS: Server-side request forgery in mcp-gateway-registryAWS Security Bulletins · Checked Nov 12, 2026
  2. GHSA-fc39-fwh3-wv5r: Upstream-address request forgerymcp-gateway-registry maintainers · Checked Nov 12, 2026
  3. mcp-gateway-registry 1.35.0 release notesmcp-gateway-registry maintainers · Checked Nov 12, 2026
  4. Server-Side Request Forgery Prevention Cheat SheetOWASP Cheat Sheet Series · Checked Nov 12, 2026
  5. Use the Instance Metadata Service to access instance metadataAmazon EC2 User Guide · Checked Nov 12, 2026

Reader discussion

What would you ask next?

Questions, useful corrections and thoughtful disagreement are welcome. Comments are reviewed before they appear.

No reader comments yet.

10–2,000 characters. Plain text. Leave out passwords, private project data and personal contact details. A blank display name appears as Anonymous.

Complete the CAPTCHA before sending. Verification runs on your device.