AI Composition Analysis
Open security intelligence for the AI component ecosystem. We index, scan, and publish risk profiles for MCP servers and Agent Skills so you can make informed decisions before adoption.
We believe skills, MCP servers, and other AI-related components are new undeterministic attack surfaces of the agentic era. AI agents rely on them from public registries — yet most have no standardized security assessment, and many introduce real risk.
Exposed tools can leak secrets, execute arbitrary code, or exfiltrate data. Registry metadata rarely includes security context.
There is no standard way to assess risk before adding an MCP server to your stack.
SKILL.md and tool definitions can contain prompt injection, data exfiltration, or unsafe syscalls. Skills are distributed across GitHub, SkillsMP, Tessl — with no central security view.
Duplicate skill names can confuse or deceive users about what they are actually running.
We run a continuous pipeline: ingest from 10+ registries, run multi-phase security scans, and publish risk profiles.
skillaudit.sh is an easy-to-run fast checker. Scan a skill or MCP before adoption — no signup, no pipeline.
We provide open, standardized vulnerability data for the AI component ecosystem — comparable to Sonatype OSS Index or Socket.dev, but for MCP servers and Agent Skills. Developers can see risk before they adopt.
CodeThreat AppSec
Full SAST + SCA agentic security analysis for MCP servers and Skills.