AI Stack Alert High CVE-2026-117569 For AppSec, AI Eng, SOC / IR, CISO AI-assisted
Kali-GPT MCP server OS command injection via unsanitized scanner parameters (CVE-2026-117569)
Kali-GPT through commit a608c74 has an OS command injection flaw in execute_tool() in kali_gpt/tools/mcp_server.py. MCP clients can pass shell metacharacters in parameters such as target, ports, domain, wordlist, or url. The values are executed via subprocess.run with shell=True as the MCP server user. The CVSS base score is 8.5.
Why it matters
Any MCP client, including a prompt-injected AI agent, can run arbitrary OS commands with the privileges of the MCP server user. Organizations that expose this server to agents or untrusted clients face host compromise.
Do this
- Identify any deployments of Kali-GPT at or before commit a608c74 and restrict or stop the MCP server until a fixed version is available.
- Check whether a fixed upstream commit or release exists and apply it. The source does not name a fixed version.
- Until patched, limit which MCP clients and agents can reach the server, and require human approval for scanner tool calls.
- Review the code path in execute_tool() in kali_gpt/tools/mcp_server.py to confirm whether subprocess.run uses shell=True with unvalidated input.
- Hunt for unexpected child processes spawned by the MCP server process, such as shells or network utilities, and review its logs for shell metacharacters in target, ports, domain, wordlist, or url fields.
Exploit Clock
Exploitation unlikely (for now)
- EPSS 0.67%
Signals
- NoCISA KEV
- 8.5CVSS
- 0.67%EPSS
- NoPublic PoC
- 34/100Priority
- 70%Confidence
Evidence
“allows MCP clients to run commands via unsanitized scanner parameters”
“executed via subprocess.run with shell=True as the MCP server user”
Sources
Affected / related
Kali-GPT, OS command injection, Kali-GPT MCP server
Published 2026-10-27 14:16 UTC · updated 2026-11-08 02:41 UTC · AI-assisted summary, checked against quoted sources · How we score