/

CVE-2026-79747

Severity:

HIGH RISK

(7.1)

Server-Side Request Forgery (CWE-918) in MCPHub API Gateway

CVE-2026-79747

Awaiting Analysis

Vulnerability type

Category:

Server-Side Request Forgery

CWE:

CWE-918

Impact:

Unauthorized access to internal services and network resources

Severity:

HIGH RISK

(7.1)

CVSS version:

3.1

Catch this CVE while code is being written

Detect vulnerable code directly in pull requests, so CVEs are fixed before merge.

Summary

MCPHub allows an authenticated non-admin user to register an MCP server at an arbitrary URL, causing the hub to make server-side requests without adequate egress restrictions. Through the OpenAPI proxy, response content may be returned to the attacker, while SSE and streamable HTTP transports can still enable blind requests to internal or otherwise restricted services.

Affected context

The issue affects MCPHub deployments before the patched release when:

  • Authenticated non-admin users can register or configure MCP servers
  • The hub accepts arbitrary destination URLs
  • Outbound requests are not filtered for loopback, RFC1918, or link-local addresses
  • A proxy or transport endpoint causes the hub to contact the registered destination

OpenAPI proxy requests may expose response bodies, whereas SSE and streamable HTTP requests may operate without returning response content.

Exploit preconditions

  • Attack surface: MCP server registration and proxy or transport endpoints
  • Required access: An authenticated account with non-administrative server-registration privileges
  • Trigger point: A request routed through a user-controlled MCP server destination
  • Why exploitation is possible: The hub performs outbound requests to attacker-selected URLs without sufficient destination or egress controls

Why this happens

The vulnerability results from insufficient validation and network enforcement around user-supplied outbound destinations.

A server-side component accepts an arbitrary URL, establishes a request to that destination, and does not reliably prevent access to loopback, private, or link-local network ranges. This behavior maps to CWE-918: Server-Side Request Forgery, because an attacker can influence a trusted server into making requests on their behalf.

Potential impact

  • Access to internal administrative or metadata services
  • Retrieval of sensitive responses through the OpenAPI proxy path
  • Blind interaction with internal services through supported transports
  • Network reconnaissance from the MCPHub host or deployment network
  • Potential follow-on compromise if an internal service trusts the hub or exposes privileged functionality

General mitigation guidance

Mitigation generally focuses on reducing trust in user-supplied destinations and enforcing outbound network policy at multiple layers.

Common defensive patterns include:

  • Allowing destinations only from an explicit, tightly governed server allowlist
  • Validating resolved IP addresses and blocking loopback, RFC1918, link-local, multicast, and other restricted ranges
  • Rechecking destinations after DNS resolution to reduce rebinding risks
  • Restricting redirects and validating every redirect target
  • Applying network egress policies outside the application as defense in depth
  • Limiting server-registration capabilities and applying least privilege to authenticated users
  • Avoiding unnecessary response reflection and minimizing sensitive error or response details

How teams detect issues like this

Organizations can identify SSRF risks through:

  • Static analysis and code review for HTTP clients using user-controlled URLs
  • Review signals involving unrestricted URL registration, redirect following, or missing IP-range validation
  • Tests covering loopback, RFC1918, link-local, encoded, redirected, and DNS-rebinding destinations
  • Egress firewall and service-mesh policies that deny restricted network ranges
  • Monitoring outbound requests from MCPHub for unusual internal targets, metadata endpoints, or unexpected destinations
  • Automated security review tools that flag server-side requests built from untrusted input

Sources

SourceLink
github.comgithub.com/samanhappy/mcphub/releases/tag/v1.0.32
github.comgithub.com/samanhappy/mcphub/security/advisories/GHSA-9wx9-prgc-gmjr

Frequently asked

questions

Yes. An attacker generally needs network access to MCPHub and a valid authenticated account with permission to register or use an MCP server. The attack does not require direct access to the targeted internal service.

No. The described issue can be triggered by an authenticated non-admin user when server registration is available to that account.

Restrict server destinations to an approved allowlist, validate resolved addresses and redirects, and enforce independent outbound network egress controls.

Treat user-provided URLs as untrusted, test all URL and DNS edge cases, minimize response reflection, and continuously monitor and restrict application egress.

Find out what's already
exploitable in your codebase.