# yotta-verify-mcp [Health: Active]

**Category:** 🔒 Security  
**Repository:** https://github.com/YottaMeta/yotta-verify-mcp  
**GitHub Stars:** 0  
**npm Downloads (last month):** 901  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/yotta-verify-mcp

## Description
Pre-install security scanner for AI agent skills (local stdio MCP). Offline static scan; no upload.

## Tools
Capabilities this server exposes over MCP:

- **scan_skill** — Pre-install scan: `target` (dir / .tgz / npm package) → verdict + severity counts + findings
- **generate_badge** — Audited badge: local SVG + shields.io URL; folds in validate / vetter / audit / version / tests
- **gate_check** — CI gate: fail when the worst severity exceeds `max_severity` (default medium)
- **get_report** — Verification report: Markdown or JSON, same format as the CLI

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

```json
"mcpServers": {
  "yotta-verify-mcp": {
    "command": "npx",
    "args": ["-y","@yottameta/yotta-verify-mcp"]
  }
}
```

## Documentation

## What yotta-verify-mcp MCP server does

The yotta-verify-mcp MCP server exposes the same scanning engine as the yotta-verify CLI through MCP. It is intended for checking an agent skill, plugin, or MCP server before installation. The scan looks for prompt-injection indicators, malicious patterns, SKILL.md integrity issues, and permissions-related findings, then returns a verdict with severity totals and individual findings.

The scanner is local and static. It reads files and produces results, but it does not execute the target, monitor it at runtime, repair detected issues, or make the final installation decision. Directory contents remain on the local machine. When the target is a public npm package, the package is downloaded to a temporary directory for scanning and then removed.

## How it works

The server uses the standard stdio MCP transport, so an MCP client starts it as a local process and discovers its four tools. `scan_skill` accepts a directory, `.tgz` or `.tar.gz` archive, or npm package name. For an npm package name, the scanner uses `npm pack` to obtain a temporary local copy before analysis.

`gate_check` applies a maximum permitted severity. Its default threshold is `medium`; the result includes whether the check passed, the observed worst severity, the configured limit, and an exit code. This makes it suitable for a pre-install or CI decision point, while still leaving the final decision to the user or workflow owner.

## Setup and configuration

The package can be installed with `npx -y @yottameta/yotta-verify-mcp`, which also starts the MCP server when no installer arguments are supplied. The repository also documents installation through `git clone`, a GitHub ZIP download, or `install.sh`. The installer can target an agent’s default skills directory with `--agent`, use an explicit location with `--dir`, or list known agent directories with `--list`.

Configuration in an MCP client is optional. An agent can present the proposed `mcpServers` entry and wait for explicit consent before changing a configuration file. If configuration is declined, the bundled CLI scanner can be used instead; the scan result is unchanged. The implementation requires Python 3.8 or newer and uses the standard library, with support stated for Windows, Linux, and macOS.

## Tools and capabilities

The yotta-verify-mcp MCP server provides these tools:

- `scan_skill` scans a local directory, archive, or npm package and returns a verdict, severity counts, and findings.
- `generate_badge` creates a local SVG badge and a shields.io URL. It can derive a verdict from a target or accept one directly, and can combine validation, vetter, audit, version, and test information.
- `gate_check` evaluates a target against `info`, `low`, `medium`, `high`, or `critical` as the maximum severity.
- `get_report` writes or returns the verification result as Markdown or JSON. Markdown is the default, and an output path is optional.

Badge version information normally refers to the bundled yotta-verify scanning engine, not necessarily the MCP package version. A custom version can be supplied when needed.

## Limitations and notes

The yotta-verify-mcp MCP server performs static analysis only. It does not provide sandboxing, dynamic analysis, runtime monitoring, remediation, or a definitive install recommendation. A clean or acceptable verdict should therefore be treated as evidence for review rather than a substitute for authorization and human judgment.

Directory scans are fully offline. Scanning an npm package requires retrieving the public package, but the target is not uploaded and its code is not run. Only evaluate targets you are authorized to inspect. The repository includes validation commands for the MCP engine and documentation consent behavior. It is released under the MIT license.

_Full upstream README: https://allmcps.com/mcp/yotta-verify-mcp/readme_