Security audit
Security checks for vulnerabilities and agentic risk
The skill’s accessibility checks and code fixes are clearly scoped, with routine precautions for external processing and optional software installation.
Hosted scans send submitted URLs and HTML to WebAbility; remove secrets and confirm private content may be shared. Review generated fixes. If choosing local execution, pin a reviewed package version or image digest and limit filesystem access.
The command npx -y @webability/mcp can download and execute an unpinned package version with the invoking user's privileges. A compromised or malicious future release could access local files or credentials; this is an optional setup path, not evidence that the current package is malicious.
The static match is incomplete: the full command specifies webabilityio/mcp:latest, so a tag is present, but it is mutable and does not pin image contents. A replaced or compromised image could execute malicious code inside the container, although the shown command does not grant privileged access or mount host directories.
The default workflow sends scan inputs to a third-party hosted MCP service without explicitly explaining that privacy boundary. Raw HTML may contain proprietary source or personal data, and URLs may contain access tokens; the rule limiting submissions to user-requested inputs reduces scope but does not ensure informed authorization for external disclosure.
npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.
Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).
Transport is Streamable HTTP.
2. Optional local option (stdio): `npx -y @webability/mcp` or `docker run -i --rm webabilityio/mcp:latest`.
3. Optional: sign up free at https://www.webability.io, create an API key, and set `WEBABILITY_API_KEY`. This unlocks the account tools `visual_audit`, `start_audit` and `get_audit`.
No suspicious patterns detected.