Skip to main content
Glama
README.md
# dnnsec

[![npm version](https://img.shields.io/npm/v/dnnsec.svg)](https://www.npmjs.com/package/dnnsec)
[![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](https://opensource.org/licenses/MIT)

**Deep Neural Network Security (DNNSEC)** MCP Server. Designed for enterprise AI systems to verify neural model weight integrity, audit adversarial robustness, and safeguard autonomous agent infrastructure.

## Features

- **Model Integrity Verification**: Cryptographic validation of neural weights and checkpoint artifacts.
- **Adversarial Robustness Screening**: Automated resilience check against FGSM/PGD perturbations.
- **Model Context Protocol (MCP)**: Native integration with Claude Desktop, Cursor, and autonomous agent frameworks.

## Quick Start

### Running directly via npx
```bash
npx dnnsec
```

### Claude Desktop Integration

Add to your `claude_desktop_config.json`:

```json
{
  "mcpServers": {
    "dnnsec": {
      "command": "npx",
      "args": ["-y", "dnnsec"]
    }
  }
}
```

## Tools Included

1. `verify_model_integrity`: Audits weights checksums and model authenticity.
2. `check_adversarial_robustness`: Evaluates resilience to input perturbation attacks.

## License

MIT © [tudadada](https://github.com/tudadada)

TDQS

B3.1/5.0

Scored across 2 tools

Disambiguation2/5

The two tools have overlapping purposes: both verify model integrity and check adversarial robustness, making it unclear when to use one versus the other. An agent would struggle to decide which tool handles a given security assessment request.

Naming Consistency4/5

Both tool names follow a consistent verb_noun pattern (verify_model_integrity, check_adversarial_robustness), though the verbs 'verify' and 'check' are near-synonyms, which slightly reduces clarity.

Tool Count2/5

With only two tools, the server feels thin for a domain as broad as neural network security. The count is at the extreme low end, offering minimal coverage for the stated purpose.

Completeness2/5

The tool surface is severely limited: it covers verification and robustness screening but lacks common operations like model hardening, attack generation, report generation, or remediation. The domain is not fully covered.

Maintenance

ActivityMaintained
ResponsivenessNo issues