Threat Intel

CVE-2026-90617: PentestAgent MCP HTTP command injection

The PentestAgent MCP HTTP task path can pass attacker-controlled input into operating-system commands.

Armor11 min read

Tracking: CVE-2026-90617
Severity: High (CVSS 3.1: 7.3); Medium under CVSS 4.0 (6.9)

Who is affected

GH05TCREW/PentestAgent through commit cf882dabea3ed91cef016cdd115e5426315665a2. Exposure depends on a caller reaching the affected MCP HTTP task interface.

How the attack works

The report identifies run_task in interface/main.py as an operating-system command-injection path. Input supplied through the MCP HTTP interface can alter the command that the host executes.

Commands run with PentestAgent's permissions. Because a penetration-testing runtime may intentionally hold network access and assessment credentials, its actual privileges are central to assessing the consequences.

What to do

  1. Restrict the affected HTTP interface to trusted callers and disable the vulnerable task path until a corrected revision is confirmed.
  2. Compare the deployed commit with the vendor's remediation references, including pull request 101. The reviewed CVE record does not establish a fixed numbered release.
  3. Review task submissions, child processes, and unexpected activity originating from the assessment host. Revoke credentials if the investigation shows exposure.

Scope and limits

The CVE record references a public proof of concept. Severity differs between the published CVSS versions; neither score establishes a confirmed compromise.

Sources

threat-intelcvecommand-injection

See what your own fleet is running

Armor1 discovers and risk-scores every MCP server, skill and tool your developers have connected. Free for personal use, no credit card.

Start Free