AI-Infra-Guard (A.I.G) provides an MCP Server along with an agent-based security scanning framework for AI infrastructure, MCP capabilities, and jailbreak evaluation. Once running, the MCP server exposes APIs and a Swagger UI that allow you to create and manage MCP scan tasks, trigger AI infra vulnerability checks, and perform Jailbreak Evaluation across supported models and components. You can leverage the built-in MCP plugin points to add or customize security checks for specific MCP Servers or Agent Skills, and inspect results through the API or the web interface. The platform is designed to be deployed via Docker and/or docker-compose, offering a modern UI and a complete API surface for integration with other tooling and CI workflows.

Prerequisites:

• Docker and Docker Compose installed on your machine
• Git and curl (optional for one-click install)

Step-by-step installation:

1. Clone the repository (or pull the official images): git clone https://github.com/Tencent/AI-Infra-Guard.git cd AI-Infra-Guard

2. Deploy using Docker Compose (preferred for quick start with pre-built images):

   If using Docker Compose v2+:

   docker-compose -f docker-compose.images.yml up -d

   If you only have docker-compose (v1) available, use the standard compose file if provided:

   docker-compose up -d

3. Alternative one-click install (auto-installs Docker and launches A.I.G): curl https://raw.githubusercontent.com/Tencent/AI-Infra-Guard/refs/heads/main/docker.sh | bash

4. Access the web UI and API:

   • Web interface: http://localhost:8088
   • API docs / Swagger: http://localhost:8088/docs/index.html

Notes:

• The project emphasizes deployment via Docker and Docker Compose. It may require network access to pull images from Docker Hub and to expose port 8088 for the UI/API.
• The repository notes that authentication is not provided in the public deployment and should not be exposed on public networks.

Tips and considerations:

• Security: The README notes that A.I.G currently lacks an authentication mechanism for public deployment. If you deploy publicly, implement network restrictions or add an authentication layer before exposing the UI/API.
• Performance: Allocate sufficient RAM and disk space per the Quick Start table (example: 4GB RAM, 10GB+ disk for Docker image deployments).
• API usage: After starting, the API docs are available at /docs/index.html for detailed parameter descriptions and example requests.
• MCP collaboration: You can extend MCP capabilities by adding new MCP plugins and datasets (fingerprints, vulnerability rules, etc.) under the data/ directories as described in the contribution guidelines.
• Versioning: Monitor the changelog for updates to new CVE coverage and new MCP scanning features, especially around new AI components.