AWS has published a security bulletin for a missing-authorization bug in Amazon mcp-gateway-registry, an open-source registry and gateway for Model Context Protocol servers. Bulletin 2026-179-AWS, dated November 11, 2026, describes CVE-2026-100307: an authenticated low-privileged user could grant themselves the registry administrator scope through a crafted group import request, then control every registered gateway asset. The issue is fixed in version 1.25.0. Upgrading is not enough on its own, because a privileged mapping written before the patch still works afterward.
Quick Take
- The flaw: The /api/servers/* management endpoints checked who the caller was, but did not check what they were allowed to do. A signed-in user could POST a crafted request to /api/servers/groups/import and map their own identity-provider group to the mcp-registry-admin scope.
- Impact: AWS and the project’s GitHub advisory say that status takes effect on the next request, with no re-login, and gives administrative control over every registered gateway asset, including the ability to read stored server credentials.
- Who is affected: mcp-gateway-registry versions 1.0.9 through 1.24.7. The project is an open-source registry used to discover, control access to, and proxy MCP servers.
- The fix: Version 1.25.0, released June 27, 2026, and documented in the November 11 bulletin, enforces authorization on the write path. After upgrading, operators still need to review the scope store, logs and credentials.
- Related issue: The same day, NVD also listed CVE-2026-125926, an SSRF in the server registration and update API before 1.35.0, fixed by upgrading to 1.35.0. Teams should not stop at 1.25.0 if they are also exposed to that bug.
In plain English
This is a bug in open-source software that some companies use as a directory and front door for AI tool servers. A person who already had a normal login could quietly make themselves an administrator, then see and change every tool hooked up to that directory, including stored passwords. A patch has been out since June.
What you should do: If you ran a vulnerable version, installing the patch is the first step, not the last, because an old quietly created admin grant can survive the upgrade. You should also move to a later release if you have not already, because a second bug in the same project was fixed later.
What happened
According to AWS and the GitHub advisory GHSA-rxf3-qgpv-wxvr, administrator status in the registry is derived per request from group mappings in a DocumentDB collection called mcp_scopes. Because the external management API authenticated the caller but skipped role checks, any authenticated user could write a privileged mapping. The advisory says a crafted import could name a privileged scope, map the caller’s group to mcp-registry-admin, or grant a mutating UI permission to all servers.
Once that mapping was stored, “the escalation took effect on the caller’s next request with no re-authentication,” AWS wrote. An attacker in that position could register, modify, toggle and remove servers and agents, read and change access-control groups, and read “the stored authentication credential and custom headers, which are decrypted on read through administrator-reachable routes.” The same missing-authorization pattern also affected sibling group routes (create, add, remove, delete) and several server-mutation routes, including register, toggle, auth-credential update, and version-default or version-delete. Those are gated in the 1.25.0 fix as well.
NVD scores CVE-2026-100307 as 8.8 (CVSS 3.1) and 8.7 (CVSS 4.0), with network access, low privileges required, and high impact on confidentiality, integrity and availability. AWS rates the bulletin “Important (requires attention).” The researcher who reported it is credited as novice-22.
What to do
The patch is 1.25.0. AWS is explicit that the upgrade “does not modify group definitions that were written before the upgrade, and it does not invalidate credentials that were read before the upgrade.” After installing it, operators who ran 1.0.9 through 1.24.7 should, in AWS’s words:
- Review mcp_scopes for group_mappings to mcp-registry-admin or another privileged scope, and for ui_permissions that grant a mutating action to all. Remove grants that were not created by an administrator.
- Search logs for POST /api/servers/groups/import from non-administrator accounts, and for the sibling group and server-mutation routes, then correlate later /api/servers/* activity from the same identity.
- Rotate registry-managed credentials if unintended access cannot be ruled out, especially if logs are incomplete.
- Inspect registered servers and agents for unexpected changes, including any upstream URL that now points at a different host.
Until an upgrade is possible, AWS suggests restricting the mutation routes at the ingress layer to trusted administrators, and watching the scope store and logs for unexpected grants. Those steps do not replace the post-upgrade review.
A separate CVE published the same day, CVE-2026-125926, describes SSRF in the server registration and update API in versions before 1.35.0, which “might allow an authenticated remote user to send requests from the gateway to internal network services and cloud instance metadata endpoints and read the responses.” The recorded fix is 1.35.0. Operators patching 100307 should check whether they also need that later release.
About the analysis
Researched and written with AI assistance under the editorial responsibility of the i10X News team. The title image is an AI-generated image. Read our AI Use Policy. Spotted an error? Request a correction.
This independent i10x analysis is based on AWS security bulletin 2026-179-AWS, GitHub advisory GHSA-rxf3-qgpv-wxvr, and the NVD records for CVE-2026-100307 and CVE-2026-125926. i10x has not exploited the issue or audited a live registry.
i10x Perspective
MCP gateways sit in front of tools that can reach internal systems, so “any logged-in user can become admin” is not a theoretical paperwork bug. The more important sentence in the AWS note is the one after the version number: the patch does not unwind a mapping that was already written. Teams that only bump the container tag and move on may still have an extra administrator in the scope store.



