CVE-2026-130750

Network-AI HTTP transport could expose tool metadata across origins

Jovancoding · Network-AI

First documented disclosure · Review checkpoint

Summary

A localhost-origin page could read metadata from a gateway bound beyond loopback.

Severity and exploitation evidence

Medium · 6.9/10 · CVSS 4.0. Severity source.

Exploitation not established: Malicious exploitation remains unverified in the reviewed originals. A demonstration or absence from KEV does not establish exploitation status.

Exploitation evidence source

Technical severity is not a measure of observed attacks or of risk to your deployment. Attempts, demonstrations, and successful compromise retain the qualifications in the source.

How the weakness works

The Streamable HTTP transport reflected localhost origins regardless of its bind interface.

Conditions required

A supported non-loopback deployment and attacker-controlled localhost-origin content, such as a local web application.

Potential impact

The unauthenticated tools and health routes reveal tool schemas and bridge metadata. MCP tool invocation still requires its bearer token.

Affected versions and configurations

Network-AI: 5.11.0 through versions before 5.17.2.

Fix and mitigation status

5.17.2.

Practical next steps

Apply the issue-specific correction: 5.17.2. Restrict the affected tool or endpoint until the correction is deployed.

Operating lesson

Apply browser-origin policy consistently across every transport.

Evidence and limits

Earliest reviewed public evidence 2026-11-02; CVE publication 2026-11-18. CVSS 4.0 6.9 is attributed to VulnCheck in the CNA record. The maintainer describes a low-impact metadata leak; the selected score is the CNA CVSS 4.0 assessment, not an inference that authenticated tools are exposed.

Sources