Attack Chain

1. An attacker supplies a user-controlled URL argument (e.g., http://[::ffff:127.0.0.1]:PORT/) to the auth_fetch or download_media tools within auth-fetch-mcp.
2. The application's navigateTo() or ctx.request.get() function calls assertSafeUrl() to validate the provided URL against SSRF protections.
3. Inside assertSafeUrl(), the isPrivateV6() function is invoked to determine if the URL's hostname is a private IPv6 address.
4. The Node.js WHATWG URL parser silently normalizes the IPv4-mapped IPv6 address (e.g., ::ffff:127.0.0.1) to its hex-normalized form (e.g., ::ffff:7f00:1).
5. isPrivateV6() attempts to check the normalized address but net.isIPv4('7f00:1') incorrectly returns false because 7f00:1 is not a dotted-decimal IPv4 string.
6. This incorrect false result causes the isPrivateV6() and subsequently assertSafeUrl() functions to treat the loopback address as safe, bypassing the intended SSRF protection.
7. The now "validated" URL, which points to an internal loopback service, is then used by page.goto() (for auth_fetch) or ctx.request.get() (for download_media) to issue an HTTP request.
8. The application fetches content from the specified internal service, extracts it, and returns it to the attacker, thereby compromising the confidentiality of internal service responses.

Impact

This Server-Side Request Forgery (SSRF) vulnerability allows an attacker who can control the url argument of auth_fetch or download_media to force the auth-fetch-mcp server to make HTTP requests to services on 127.0.0.1 or other private IPv4 ranges. This impacts end users running auth-fetch-mcp locally, where an attacker could read responses from local development servers, admin panels, or credential endpoints. Server-side deployments of auth-fetch-mcp face the same risk against internal network services. The confidentiality of internal service responses is fully compromised (C:H), as demonstrated by the ability to retrieve an INTERNAL_SECRET_MARKER from a simulated internal service. The integrity and availability of the target service are not directly affected.

Recommendation

• Patch CVE-2026-49857: Implement the proposed remediation outlined in the GHSA advisory, which involves decoding the hex-encoded IPv4-mapped suffix before passing it to isPrivateV4() in src/security.ts.
• Application-level Guard: Add a BrowserContext route guard in src/browser.ts to re-validate every navigation URL, including redirect targets, using assertSafeUrl() for enhanced protection.
• Review auth-fetch-mcp usage: If auth-fetch-mcp is exposed to user-controlled input, review configurations and consider restricting its ability to resolve arbitrary URLs until a patch for CVE-2026-49857 is available.