New from Wiz Research: Critical NGINX flaw. Understand the risk and impact.
Vulnerability DatabaseGHSA-38j4-j742-j7r3

GHSA-38j4-j742-j7r3: 
JavaScript vulnerability analysis and mitigation

Summary

This is a precedent-consistency report, not a novel class: the identical NAT64 well-known-prefix gap in the sibling fetch server's SSRF guard was accepted by this repository, published as GHSA-66qx-2266-pmj6, and fixed in v2.1.0 (#5238). The everything server's gzip fetch guard (tools/gzip-file-as-resource.ts) still classifies addresses in 64:ff9b::/96 as public, because the prefix is never unwrapped to the embedded IPv4 before the block-list check. Impact is contingent on a NAT64/DNS64 path that can actually reach the embedded IPv4; on ordinary networks the connection simply fails. The code comment in the guard already acknowledges the gap (disclosed below).

Details

Affected: Product @modelcontextprotocol/server-everything (modelcontextprotocol/servers, src/everything). Confirmed in monorepo v2/main at commit 6cfff622fb8eb41e2504340d5a6105d34094143c (2026-11-11). Precedent: mcp-server-fetch in this same repository — GHSA-66qx-2266-pmj6 (originally reported as issue #5181), fixed in v2.1.0 (2026-10-24) via #5238 and tracked by maintainer in #5255. Other versions not verified. Details:

  • src/everything/tools/gzip-file-as-resource.ts (line numbers at pinned commit): guard's block list registers the local-use NAT64 range (64:ff9b:1::/48, L96) but not the well-known prefix; assertPermittedTarget (L285-311) checks every address returned by lookup(host, { all: true }) (L302) against that block list, so an address of form 64:ff9b::<embedded IPv4> is always classified as public.
  • Disclosure of in-code self-admission: comment at L70-72 already states, "The well-known NAT64 prefix (64:ff9b::/96) is not unwrapped and so passes as public." This report does not present the gap as newly discovered in principle; it reports that an acknowledged gap in a comment remains unfixed in code while the identical gap in the sibling guard was treated as a vulnerability, published, and fixed. (Unlike fetch's case, nothing in the everything README or features documentation mentions NAT64 to users.)
  • Mechanism: (1) a client passes gzip-file-as-resource tool a URL such as http://[64:ff9b::a9fe:a9fe]/ (embedding of 169.254.169.254); for an IP-literal host the resolver returns the address itself; (2) allowlist and private-address checks pass, because checked address is NAT64 form and classified public; (3) fetch is attempted, and where a NAT64 translator sits on the path, connection is translated to embedded IPv4. In DNS64 environments an ordinary hostname can also produce this address form via synthesized AAAA records.
  • This is NOT the DNS-rebinding window the README documents for this guard: there, a name re-resolves between check and connect. Here the address checked is itself the NAT64 form and is misclassified at check time; translation happens in network, not by re-resolution. The documented limitation is therefore not claimed as part of this report.

Proof of concept

Established by code inspection against pinned tree. Classification outcome follows deterministically from guard's contents: with 64:ff9b::/96 absent from every registered non-public range, assertPermittedTarget accepts a target whose resolved address lies in that prefix, including embeddings of RFC 1918 and link-local IPv4 addresses. A unit-level exercise of the classifier mirrors the proof accepted for the fetch precedent. End-to-end connection was NOT dynamically demonstrated: no NAT64/DNS64 environment was available, and on a network without one the connection attempt simply fails. Appendix regression test (ev-k03-1_gzip-nat64-classification.test.ts, style of existing tests/gzip-private-addresses.test.ts): pins NAT64 address judged by wrapped IPv4 — embeddings of private/loopback/link-local refused, public embedding not refused for prefix alone. Verification (Node v24.22.1 / vitest 4.1.10, pinned tree): test FAILS on pinned code — 5 of 7 cases fail, tool fetched NAT64 target and returned resource_link instead of refusal (2 public controls pass). With remediation on scratch copy (unwrap 64:ff9b::/96 and classify embedded IPv4, mirroring #5238), test PASSES 7/7, existing gzip suites stay green 41/41. Test exercises classification only — DNS and fetch stubbed, no real NAT64 gateway. Full test source in reporter's evidence files, available on request.

Impact

Only in deployments where everything server runs with a NAT64/DNS64 path to internal address space: a client able to call gzip fetch tool can make server connect to embedded IPv4 targets — including private and link-local addresses — that guard is documented to refuse by default. No impact on ordinary networks. Remediation: Mirror fetch fix (#5238) in gzip guard: unwrap 64:ff9b::/96 before classification and subject embedded IPv4 to same block-list check (local-use range 64:ff9b:1::/48 is already registered; well-known prefix needs unwrap-and-check rather than flat block, so public embedded addresses keep working where NAT64 is intended path). Honest limitations:

  • Class novelty is zero: same defect class reported by third party (#5181), published (GHSA-66qx-2266-pmj6), fixed for fetch. Contribution is limited to identifying surviving instance in second guard implementation.
  • Exploitation contingent on NAT64/DNS64 infrastructure; reporter could not and did not demonstrate a completed connection.
  • No testing against any third-party or production deployment. Reporter credit: Tomoki Murota, CISSP

Source: NVD

Related JavaScript vulnerabilities:

CVE ID

Severity

Score

Technologies

Component name

CISA KEV exploit

Has fix

Published date

GHSA-5vm7-vc2x-p3w2HIGH7.5
  • JavaScript logoJavaScript
  • graphql-upload
NoNoNov 18, 2026
GHSA-p4q5-rhgr-hg46HIGH7.5
  • JavaScript logoJavaScript
  • @orpc/server
NoYesNov 18, 2026
GHSA-3xgv-mvj6-g78rMEDIUM6.5
  • JavaScript logoJavaScript
  • @modelcontextprotocol/inspector
NoYesNov 18, 2026
GHSA-qwwm-252j-9whwMEDIUM5.3
  • JavaScript logoJavaScript
  • graphql-upload
NoNoNov 18, 2026
GHSA-38j4-j742-j7r3LOW3.7
  • JavaScript logoJavaScript
  • @modelcontextprotocol/server-everything
NoYesNov 18, 2026

Free Vulnerability Assessment

Benchmark your Cloud Security Posture

Evaluate your cloud security practices across 9 security domains to benchmark your risk level and identify gaps in your defenses.

Request assessment

Additional Wiz resources

Get a personalized demo

Ready to see Wiz in action?

"Best User Experience I have ever seen, provides full visibility to cloud workloads."
David EstlickCISO
"Wiz provides a single pane of glass to see what is going on in our cloud environments."
Adam FletcherChief Security Officer
"We know that if Wiz identifies something as critical, it actually is."
Greg PoniatowskiHead of Threat and Vulnerability Management