AI Stack Alert High CVE-2026-117568 For AppSec, AI Eng, SOC / IR, CISO AI-assisted
CVE-2026-117568: OS command injection in mcp-ffmpeg through 1.0.5
mcp-ffmpeg through version 1.0.5 has an OS command injection flaw. Crafted videoPath or outputDir values sent to the resize-video, extract-audio and get-video-info tools can run shell commands. The flaw is in the child_process.exec command string, and commands run as the MCP server user. The CVSS base score is 7.3.
Why it matters
Any client that can call these MCP tools, or any party who can place a file or directory with a crafted name where the server reads or writes, may execute commands as the MCP server user. This exposes the host running the server and any data or credentials it can reach.
Do this
- Identify all deployments of mcp-ffmpeg and confirm the installed version. Treat 1.0.5 and earlier as affected.
- Update mcp-ffmpeg to a fixed release once the maintainer publishes one. Check the upstream project and NVD for the fixed version.
- Until patched, restrict which clients can call resize-video, extract-audio and get-video-info, and disable these tools if they are not needed.
- Limit the MCP server user's filesystem access and network reach, and run the server in a container or sandbox.
- Audit directories the server reads or writes for file or directory names containing $(), backticks or double quotes, and review server logs for matching tool calls.
Exploit Clock
Exploitation unlikely (for now)
- EPSS 0.60%
Signals
- NoCISA KEV
- 7.3CVSS
- 0.60%EPSS
- NoPublic PoC
- 34/100Priority
- 70%Confidence
Evidence
“mcp-ffmpeg through 1.0.5 contains an OS command injection vulnerability that allows attackers to run shell commands by supplying crafted videoPath or outputDir values”
“Attackers can embed $(), backticks or a closing double quote in an existing file or directory name to break out of the child_process.exec command string”
Sources
Affected / related
mcp-ffmpeg, mcp-ffmpeg, OS command injection
Published 2026-10-27 14:16 UTC · updated 2026-11-08 02:41 UTC · AI-assisted summary, checked against quoted sources · How we score