# How to Find Example Use Cases for AI-Infra-Guard: A Complete Guide

> Discover example use cases for AI-Infra-Guard in the official Tencent repository. Learn about deployment, vulnerability scanning, MCP analysis, and more.

- Repository: [Tencent/AI-Infra-Guard](https://github.com/tencent/AI-Infra-Guard)
- Tags: how-to-guide
- Published: 2026-08-26

---

**Example use cases for AI-Infra-Guard are documented in the repository's top-level README.md, specialized submodule guides (mcp-scan, agent-scan, skill-scan), and the docs/ directory, covering Docker deployment, AI infrastructure vulnerability scanning, MCP server analysis, agent workflow testing, and jailbreak evaluation.**

AI-Infra-Guard (A.I.G) is a modular AI red-team platform developed by Tencent that bundles four main security engines into a unified testing framework. Whether you are scanning local vLLM instances for CVE vulnerabilities, analyzing MCP servers for unsafe tool hijacking, or evaluating LLM jailbreak resistance, concrete implementation examples are embedded directly in the source repository. This guide maps each primary use case to its exact file location and provides copy-paste commands to reproduce real-world security testing scenarios.

## Core Engine Capabilities and Example Applications

### AI Infrastructure Vulnerability Scanning

The **AI-infra vulnerability scan** engine fingerprints running AI services such as **vLLM**, **Ollama**, and **ComfyUI**, matching them against more than 2,000 CVE entries. A typical use case involves scanning a locally running vLLM endpoint to discover outdated model libraries or dangerous misconfigurations before production deployment.

### MCP and Skill Scanner Analysis

The **MCP & Skill scanner** analyzes MCP servers and agent-skill bundles using YAML fingerprints and CVE rules to detect unsafe tool hijacking or privilege escalation vulnerabilities. Users can upload a GitHub URL of a custom MCP server or analyze a zipped skill project without manually cloning repositories.

### Multi-Agent Workflow Testing

The **Agent-scan** engine executes multi-agent workflows (Dify, Coze, or custom implementations) to check for loss-of-control, privilege-escalation, and data-leakage vulnerabilities. This engine is essential for validating that production AI agent pipelines cannot be manipulated into unauthorized actions.

### Jailbreak Resistance Evaluation

The **Jailbreak evaluation** engine runs curated datasets against configured LLM endpoints, reporting success rates per attack type such as **"Many-Shot"** prompting. Security teams use this to assess the robustness of fine-tuned models against adversarial inputs before public release.

## Step-by-Step Implementation Examples

### Quick Docker Deployment for Web UI

To launch the complete AI-Infra-Guard platform with a pre-built web interface, execute the following commands from the repository root as documented in [`README.md`](https://github.com/Tencent/AI-Infra-Guard/blob/main/README.md):

```bash
git clone https://github.com/Tencent/AI-Infra-Guard.git
cd AI-Infra-Guard
docker-compose -f docker-compose.images.yml up -d

```

After deployment, the UI is reachable at `http://localhost:8088` where you can access all scanning engines through a graphical interface.

### Scanning a Running vLLM Service

To perform an AI infrastructure scan against a local vLLM endpoint, navigate to the web UI and select "AI基础设施安全扫描" (AI Infra Security Scan). Enter the target URL:

```bash
http://127.0.0.1:8000

```

Click "Start Scan" to fingerprint the service and check against the CVE database. This example is detailed in the [`README.md`](https://github.com/Tencent/AI-Infra-Guard/blob/main/README.md) Quick Usage Guide section.

### Using the aig-skill-scan CLI

For standalone skill analysis, install the dedicated CLI tool and scan local skill directories as shown in [`skills/aig-scanner/README.md`](https://github.com/Tencent/AI-Infra-Guard/blob/main/skills/aig-scanner/README.md):

```bash
pip install aig-skill-scan
export LLM_API_KEY="your-api-key"
aig-skill-scan --repo /path/to/skill \
               -m deepseek-v4-flash \
               --language en \
               -o result.json

```

### Remote MCP Server Analysis

To analyze an MCP server directly from a remote GitHub repository without manually cloning, use the `mcp-scan` CLI as documented in [`mcp-scan/README.md`](https://github.com/Tencent/AI-Infra-Guard/blob/main/mcp-scan/README.md):

```bash
./ai-infra-guard mcp-scan https://github.com/yourorg/your-mcp-server

```

The CLI automatically pulls the repository and executes the security scanner against the MCP configuration.

### Jailbreak Evaluation Configuration

To evaluate a custom LLM endpoint against jailbreak datasets, first configure the model in the UI under **Settings → Model Config**:

```bash
Base URL: http://127.0.0.1:8000/v1
API Key: sk-your-key

```

Then select a dataset (e.g., "Many-Shot") and click "Start Evaluation" to generate quantitative resistance metrics. Sample test cases are available in the `data/eval/example-datasets` directory.

### OpenClaw Skill Integration

Integrate AI-Infra-Guard as an OpenClaw skill to enable security scanning from chat interfaces, as documented in [`skills/aig-scanner/README.md`](https://github.com/Tencent/AI-Infra-Guard/blob/main/skills/aig-scanner/README.md):

```bash
clawhub install aig-scanner
export AIG_BASE_URL=https://aig.example.com/

```

Once installed, you can invoke scanning capabilities directly from any OpenClaw chat session.

## Key Documentation Files and Source Locations

Understanding the repository structure helps locate specific examples quickly. According to the Tencent/AI-Infra-Guard source code, the following files contain authoritative use-case documentation:

- **[`README.md`](https://github.com/Tencent/AI-Infra-Guard/blob/main/README.md)** (top-level): Central hub for Docker deployment commands, CLI snippets, and UI quick-start guides.
- **[`docs/architecture_evolution.md`](https://github.com/Tencent/AI-Infra-Guard/blob/main/docs/architecture_evolution.md)**: Illustrates how the four engines interconnect, helping you map specific use cases to the correct component.
- **[`mcp-scan/README.md`](https://github.com/Tencent/AI-Infra-Guard/blob/main/mcp-scan/README.md)**: Details remote URL scanning syntax, YAML configuration files, and example prompts for MCP security analysis.
- **[`agent-scan/README.md`](https://github.com/Tencent/AI-Infra-Guard/blob/main/agent-scan/README.md)**: Contains CLI instructions for feeding Dify or Coze workflow definitions to the agent-scanner and interpreting multi-agent test results.
- **[`skills/aig-scanner/README.md`](https://github.com/Tencent/AI-Infra-Guard/blob/main/skills/aig-scanner/README.md)**: Provides exact CLI syntax for the skill-scan tool and OpenClaw integration steps.
- **[`api.md`](https://github.com/Tencent/AI-Infra-Guard/blob/main/api.md)**: Full Swagger-compatible specification for programmatic API access to all scanning engines.
- **[`CHANGELOG.md`](https://github.com/Tencent/AI-Infra-Guard/blob/main/CHANGELOG.md)**: Lists new features (such as SkillJack additions or new jailbreak datasets) that inspire fresh test scenarios.
- **`data/eval/example-datasets/`**: Directory containing ready-made jailbreak test cases demonstrating real-world evaluation usage.

## Summary

- **Example use cases for AI-Infra-Guard** are primarily documented in the top-level [`README.md`](https://github.com/Tencent/AI-Infra-Guard/blob/main/README.md) and specialized submodule guides under `mcp-scan/`, `agent-scan/`, and `skills/`.
- The platform provides four core engines: **AI infrastructure scanning**, **MCP/skill analysis**, **multi-agent workflow testing**, and **jailbreak evaluation**.
- Docker deployment via [`docker-compose.images.yml`](https://github.com/Tencent/AI-Infra-Guard/blob/main/docker-compose.images.yml) provides immediate access to the web UI at `localhost:8088`.
- CLI tools like `aig-skill-scan` and `mcp-scan` enable headless automation and CI/CD integration for security pipelines.
- Sample datasets in `data/eval/example-datasets/` provide concrete jailbreak test scenarios for immediate LLM evaluation.
- Integration examples for OpenClaw (`aig-scanner` skill) demonstrate how to embed security scanning into existing agent workflows.

## Frequently Asked Questions

### Where are the primary example commands documented in AI-Infra-Guard?

The primary example commands are documented in the top-level [`README.md`](https://github.com/Tencent/AI-Infra-Guard/blob/main/README.md) file, which includes Docker startup instructions, CLI usage snippets, and UI configuration steps. Additional engine-specific examples are located in the respective submodule README files such as [`mcp-scan/README.md`](https://github.com/Tencent/AI-Infra-Guard/blob/main/mcp-scan/README.md) and [`agent-scan/README.md`](https://github.com/Tencent/AI-Infra-Guard/blob/main/agent-scan/README.md).

### Can I run AI-Infra-Guard without Docker for testing individual components?

Yes. You can install individual components like `aig-skill-scan` via pip and run CLI commands directly against local repositories or remote URLs. The `mcp-scan` and `agent-scan` modules also provide standalone binaries that execute without requiring the full Docker stack.

### How do I test my custom LLM model against jailbreak attacks using AI-Infra-Guard?

Configure your model endpoint in the web UI under **Settings → Model Config** by providing the base URL and API key. Then navigate to the Jailbreak Evaluation section, select a dataset from `data/eval/example-datasets` (such as "Many-Shot"), and execute the evaluation to receive quantitative resistance metrics per attack type.

### What file contains the API specification for integrating AI-Infra-Guard into my own platform?

The [`api.md`](https://github.com/Tencent/AI-Infra-Guard/blob/main/api.md) file at the repository root contains the complete Swagger-compatible API specification. This document details all programmatic endpoints for triggering scans, retrieving results, and managing configurations, enabling seamless integration with external security platforms and automated testing pipelines.