/

CVE-2026-54549

Severity:

HIGH RISK

(8.3)

SSRF (CWE-918) in Meta Ads MCP Image Upload Server

CVE-2026-54549

Deferred

Vulnerability type

Category:

Server-Side Request Forgery

CWE:

CWE-918

Impact:

Internal Network Access and Sensitive Data Exposure

Severity:

HIGH RISK

(8.3)

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

A server-side request forgery vulnerability in Meta Ads MCP allows a network caller to make the server request attacker-selected URLs through the image upload function. Because redirects, URL schemes, destinations, and resolved IP addresses are not sufficiently restricted, the server may reach internal services, cloud metadata endpoints, or other protected network locations.

Affected context

The issue occurs in streamable-HTTP deployments where the image upload tool accepts a caller-controlled image URL and downloads it before Meta credential validation completes. Deployments are most exposed when the server can reach loopback interfaces, private network ranges, cloud metadata services, or internal applications that are not intended to accept requests from the MCP process.

Exploit preconditions

  • Attack surface: The upload_ad_image tool exposed through a streamable-HTTP MCP deployment
  • Required access: Network access to the MCP server and any non-empty authorization value accepted before Meta credential validation
  • Trigger point: Server-side image retrieval from the supplied image URL
  • Why exploitation is possible: The HTTP client follows redirects without adequately validating the URL scheme, destination host, or resolved IP address, allowing requests to internal or otherwise restricted targets

Why this happens

The underlying bug is insufficient validation of attacker-controlled URLs before server-side retrieval. The downloader permits broad destinations and follows redirects without reapplying effective scheme, hostname, and IP-range restrictions. This behavior maps to CWE-918, where an application is induced to send requests from a trusted server context to unintended internal or protected resources.

Potential impact

  • Internal service responses could be exposed to the caller
  • Cloud instance metadata or other sensitive configuration could be disclosed
  • State-changing requests could be sent to reachable internal services
  • Internal applications could be probed or abused through the MCP server
  • Reachable services could experience disruption or resource consumption
  • Network segmentation assumptions could be weakened

General mitigation guidance

Mitigation generally focuses on treating remote image URLs as untrusted and

enforcing destination controls before and during retrieval.

Common defensive patterns include:

  • Allowing only explicitly approved URL schemes, hosts, and ports
  • Resolving hostnames and rejecting loopback, link-local, private, reserved, and otherwise prohibited IP ranges
  • Rechecking every redirect target rather than trusting the initial URL
  • Disabling redirects by default or permitting redirects only within an approved destination set
  • Applying strict connection, response-size, and download-time limits
  • Performing authentication and authorization before any network fetch
  • Using an isolated egress proxy or restricted network identity for media retrieval
  • Granting the MCP process only the network and service privileges it requires

How teams detect issues like this

Organizations can identify SSRF risks through:

  • Static analysis for HTTP clients consuming user-controlled URLs
  • Code review checks for unrestricted redirects and missing scheme, host, port, or IP validation
  • Tests covering loopback, private, link-local, metadata, alternate numeric, and redirect-based destinations
  • Egress policies that alert on attempts to reach prohibited address ranges
  • Network telemetry correlating MCP requests with unusual internal destinations
  • Policy enforcement requiring centralized, approved URL-fetching utilities

Automated security review tools may help flag server-side request flows, but

findings should be validated against deployment-specific network boundaries.

Sources

SourceLink
github.comgithub.com/pipeboard-co/meta-ads-mcp/commit/7d9926336bbdac6285a988d043c4ccfe126c94c5
github.comgithub.com/pipeboard-co/meta-ads-mcp/pull/138
github.comgithub.com/pipeboard-co/meta-ads-mcp/releases/tag/1.0.115
github.comgithub.com/pipeboard-co/meta-ads-mcp/security/advisories/GHSA-45gf-fjxp-cjpq

Frequently asked

questions

In a streamable-HTTP deployment, a network caller may trigger the issue if the

request reaches the MCP server and the deployment accepts the relevant

authorization value before downstream credential validation.

The caller generally needs network access to the MCP endpoint and must be able

to invoke the image upload functionality. The server must also have network

reachability to the intended internal destination.

Validate URLs before fetching them, restrict schemes and destinations with an

allowlist, reject prohibited resolved IP ranges, and revalidate every

redirect. Authorization should occur before any outbound request.

Centralize outbound URL fetching, enforce egress controls, test redirect and

DNS-resolution edge cases, monitor internal-boundary traffic, and run the

service with least-privilege network access.

Find out what's already
exploitable in your codebase.