AI Security Scanner for the components your agents trust
Paste a repo URL or an npm:/pypi: package and get an evidence-backed verdict — malicious, high-risk, or clean — before you hand an AI component the keys to your files and shell. Free, no account, the engine is open source.
Scan a component — free →What it scans
MCP servers, agent skills and plugins, and npm or PyPI packages — anything you would wire into Claude, Cursor, or your own agent. Analysis is derived only from the component itself: its code, manifests, and instruction files.
How the verdict works
Detection is deterministic — regex and AST, never an LLM, and your code is never executed. Every confirmed finding is anchored to a file, line, and snippet. The report separates malicious indicators (deliberate deception: tool poisoning, hidden Unicode, prompt injection, decode-and-execute) from powerful capabilities (often legitimate: filesystem, network, shell). Capability is not the same as risk, so the scanner does not cry wolf.
Scan by ecosystem
- MCP server security — dangerous tools and tool poisoning.
- npm package security — install scripts and obfuscated droppers.
- PyPI package security — import-time execution and RAT droppers.
- GitHub repo scanner — scan any repo, branch, subfolder or commit.
- How it works — the methodology behind every verdict.
FAQ
- Is it free?
- Yes. The full static report is free and needs no account. The engine is open source (Apache-2.0) and also runs locally via pipx install skilltotal.
- Do you run or upload my code?
- No. Detection is static (regex + AST); your code is never executed and the engine never calls an LLM. The website runs the same engine for you on the component you submit.
- What can it detect?
- Install-time execution, decode-and-execute droppers, credential/secret exfiltration, dangerous MCP tools and tool poisoning, prompt injection (including homoglyph/zero-width obfuscation), and sensitive-path access — each with file:line evidence.
- What do I get?
- A risk score out of 100, a capability breakdown, findings with evidence, and JSON/SARIF export for your own pipeline.