Vulnerability  ·  2026-08-25

MCP FFmpeg Helper server — OS command injection via unsanitized format argument

VulnerabilityMedium impactGlobalCVE-2026-78430
NVD published CVE-2026-78430 (CVSS 5.3 Medium) on 2026-08-24, describing OS command injection in the Tool Handler component of the mcp-ffmpeg-helper MCP server via the 'format' argument.
It is a concrete example of the broader MCP-server command-injection pattern where LLM-agent-callable tools pass unsanitized parameters into shell commands, but the narrow single-maintainer deployment and local-attack requirement limit real-world blast radius.
The handleToolCall function in src/tools/handlers.ts fails to sanitize the 'format' argument before passing it to an underlying FFmpeg command execution, allowing OS command injection when a local attacker (or a prompt-injected agent) supplies a crafted format value.
sworddut mcp-ffmpeg-helper 0.1.0, 0.1.1, 0.2.1
No patched version listed as of publication; avoid exposing this MCP server to untrusted input and sanitize/allow-list the format parameter as a workaround.
NVD - CVE-2026-78430GitHub - sworddut/mcp-ffmpeg-helper
See this in the live feed Explore related AI security and governance findings — updated every morning.
Open the feed →