/

CVE-2026-53957

Severity:

HIGH RISK

(7.7)

Server-Side Request Forgery (CWE-918) in Contentful MCP

CVE-2026-53957

Deferred

Vulnerability type

Category:

Server-Side Request Forgery

CWE:

CWE-918

Impact:

Credential Disclosure and Unauthorized API Access

Severity:

HIGH RISK

(7.7)

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 (SSRF) vulnerability in Contentful MCP migration tools allows LLM-controlled arguments to influence the destination of Contentful Management API requests. If migration tools are enabled and an attacker can trigger a direct MCP call or influence Contentful content processed by the server, the server may forward its management token to an attacker-controlled host or proxy.

Affected context

The issue occurs in deployments that enable the space-to-space migration tools and allow the export_space or import_space operations to process host, proxy, rawProxy, or insecure network options from tool arguments. The risk is greater when attacker-controlled Contentful content can influence model behavior or when untrusted callers can invoke MCP tools directly. Regular tools that pin the API host through server configuration are not affected by this behavior.

Exploit preconditions

  • Attack surface: Contentful MCP migration tools and their network request configuration
  • Required access: Ability to invoke an enabled migration tool directly or influence content and prompts processed by the MCP-connected model
  • Trigger point: Execution of export_space or import_space with attacker-influenced network options
  • Why exploitation is possible: User-controlled destination or proxy settings are combined with the server's management token, allowing authenticated requests to be redirected outside the intended Contentful endpoint

Why this happens

The vulnerability results from treating LLM-controlled tool arguments as trusted network configuration. Migration tools accept host and proxy-related options and use them when constructing Management API requests, while also attaching the server-held authorization token. This maps to CWE-918: Server-Side Request Forgery, because attacker-influenced input can cause the server to make requests to an unintended destination and disclose credentials through those requests.

Potential impact

  • The Contentful management token could be disclosed to an attacker-controlled host or proxy.
  • An attacker could obtain persistent API access to Contentful spaces within the token's scope.
  • Contentful content, environments, or space configuration could be modified or deleted.
  • Confidential management API responses could be exposed.
  • A compromised token could support wider data tampering or supply-chain abuse.

General mitigation guidance

Mitigation typically focuses on enforcing destination integrity and reducing the authority of credentials used by model-accessible tools.

Common defensive patterns include:

  • Pinning API hosts and proxy destinations to trusted server-side configuration.
  • Ignoring or rejecting host, proxy, rawProxy, and insecure network options supplied through LLM-controlled arguments.
  • Separating tool arguments from authentication and transport configuration.
  • Applying strict allowlists for outbound hosts, schemes, ports, and redirect targets.
  • Using narrowly scoped, short-lived credentials where supported and avoiding broad personal access tokens for automated tools.
  • Requiring explicit authorization for migration operations and isolating tools that can change space content.
  • Treating external content as untrusted input and limiting prompt-injection influence over security-sensitive tool parameters.

How teams detect issues like this

Organizations commonly detect SSRF and credential-routing risks through:

  • Static analysis searches for user- or model-controlled values passed into HTTP client hosts, proxies, URL constructors, or redirect handlers.
  • Code review checks ensuring authentication headers are never attached to requests whose destination is selected by untrusted input.
  • Policy enforcement that restricts outbound connections from MCP services to approved Contentful endpoints.
  • Runtime monitoring for unexpected DNS lookups, outbound requests, proxy usage, redirects, or authorization headers sent outside approved domains.
  • Tests that attempt to supply alternate hosts and proxy settings to every model-accessible network tool.

Automated security review tools can help identify tainted data flows from tool arguments into request destinations.

Sources

SourceLink
github.comgithub.com/contentful/contentful-mcp-server/commit/fa7477ee48515f4248bc91a025eab0ca83423fe0
github.comgithub.com/contentful/contentful-mcp-server/pull/376
github.comgithub.com/contentful/contentful-mcp-server/releases/tag/mcp-server@1.7.19
github.comgithub.com/contentful/contentful-mcp-server/releases/tag/mcp-tools@0.4.5
github.comgithub.com/contentful/contentful-mcp-server/security/advisories/GHSA-2xhg-73j7-rrgx

Frequently asked

questions

Potentially. Exploitation depends on whether an untrusted user can invoke the migration tools or influence attacker-controlled Contentful content processed by the connected model.

No. The server supplies its configured management token when constructing the request, so the core risk is redirecting that authenticated request to an attacker-controlled destination.

Keep hosts, proxies, transport security, and authentication entirely under trusted server-side control. Reject model-controlled network destinations and allow outbound requests only to explicitly approved endpoints.

Treat tool arguments and retrieved content as untrusted, use least-privilege credentials, enforce outbound network allowlists, and test every model-accessible tool for tainted URL or proxy configuration.

Find out what's already
exploitable in your codebase.