# SkillTotal [Health: Active]

**Category:** 💻 Developer Tools  
**Repository:** https://github.com/pezhik/skilltotal  
**GitHub Stars:** 12  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/skilltotal

## Description
Deterministic security scan of MCP servers, agent skills and npm/PyPI packages. Runs locally.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `uvx` (confidence: high):

```json
"mcpServers": {
  "skilltotal": {
    "command": "uvx",
    "args": ["skilltotal"]
  }
}
```

## Documentation

## What SkillTotal MCP server does

SkillTotal MCP server scans AI-related components for supply-chain risks, dangerous capabilities, prompt-injection surfaces, and possible data-exfiltration paths. Inputs can be local directories, Git repositories, ZIP or tar.gz archives, single files, npm packages, and PyPI packages. The engine examines the component itself rather than the surrounding user, organization, deployment, or runtime environment.

The static report records confirmed findings with the file, line range, and code excerpt that support each result. Signals that cannot be supported by evidence are placed in `needs_review` and do not affect the score. Reports can include behavioral trait fingerprints mapped to CSA agentic threats, MAESTRO layers, and MITRE ATLAS tactics, along with an execution-context view covering embedded credentials, delegated OAuth/OIDC, and scoped identities.

## How it works

SkillTotal MCP server runs locally with pure Python standard-library code and no runtime dependencies. Its static engine uses deterministic regular-expression and AST analysis; it does not run the target component and does not use an LLM. The same input produces the same result, which makes reports suitable for repeatable reviews and CI checks.

The `scan` command produces human-readable output by default, JSON on request, or SARIF 2.1.0 for GitHub Code Scanning and compatible IDE workflows. A scan can fail with exit code 2 when a severity threshold or risk-score threshold is reached. Baselines suppress previously accepted findings using a stable rule, file, and snippet fingerprint rather than line numbers.

`diff` compares two sources or saved reports and identifies new, resolved, and changed findings, evidence changes, capability changes, and score differences. Its `--fail-on-new` option gates only on risk introduced by the newer version. `guard` provides a pre-install allow or block decision, while `inventory` discovers configured AI components and can scan them or produce a CycloneDX 1.6 AI-BOM.

## Setup and configuration

Install Python 3.10 or newer, then install the package with `pipx install skilltotal` or `pip install skilltotal`. Git is needed only when scanning remote URLs. Source development installation is available with `pip install -e ".[dev]"`.

Typical commands include `skilltotal scan ./component`, `skilltotal scan npm:package`, and `skilltotal scan pypi:package`. Use `--json` or `--sarif --output report.sarif` for machine-readable results. Repeated `--exclude` options skip paths, and `--provenance` adds optional npm or PyPI registry metadata to `needs_review` without scoring it.

## Tools and capabilities

- Scan local, remote, archive, file, npm, and PyPI sources.
- Emit human-readable, JSON, and SARIF reports.
- Enforce severity or score thresholds in CI.
- Create baselines and compare component versions.
- Guard installations and inspect installed AI components.
- List available detection rules.

## Limitations and notes

SkillTotal MCP server analyzes files only; it does not assess runtime behavior, deployment context, or the scanner user's environment. The README describes optional dynamic analysis as a separate paid service running in an isolated sandbox with consent, not as part of the local static engine. Capabilities such as shell or network access do not automatically block a component; blocking is based on malicious indicators or configured risk thresholds. The local scanner is offline, while registry provenance is an opt-in metadata feature for npm and PyPI sources.

_Full upstream README: https://allmcps.com/mcp/skilltotal/readme_