AI-Infra-Guard (A.I.G) is a full-stack AI red teaming platform from Tencent Zhuque Lab that scans agents, MCP servers, agent skills and AI infrastructure for security risk and evaluates large language models against jailbreak attacks, built for security engineers and AI platform teams that need to self-examine their own AI deployments.

What it is

AI-Infra-Guard is an open-source Python project, licensed under Apache-2.0 and published by Tencent Zhuque Lab, that bundles several scanners into one platform: ClawScan (OpenClaw Security Scan), Agent Scan, AI infrastructure vulnerability scan, MCP Server and Agent Skills scan, and Jailbreak Evaluation. It ships with a vulnerability library covering 146 AI components and more than 2000 CVE rules, and its frontend was fully open-sourced in the v4.5.0 release. Documentation and the project homepage are hosted at tencent.github.io/AI-Infra-Guard, with translations available in Chinese, Japanese, Spanish, German, French, Korean, Portuguese and Russian.

The concrete problem it addresses is the absence of a single self-examination tool for the AI stack. Teams deploying agents, MCP servers and skill packages previously had to assemble separate utilities for each layer. AI-Infra-Guard replaces that ad hoc collection with one platform, and since v4.5.0 it also ships Skill scan, MCP scan and Agent scan as standalone CLIs, so the same engines can run outside the web interface. Later releases extend this to LLM API poisoning detection, a black-box audit of model substitution and backdoor risk, and an API Checker that fingerprints models including Gemini, Gemma and GLM families.

Key capabilities

• Scans MCP servers and agent skills, with MCP-Scan preventing remote code execution through tool whitelisting in dynamic mode.
• Runs Agent Scan against a mutation engine refactored in v5.0.0, covering OWASP skills and web-exfiltration detection across 10 skills.
• Detects skill-level bypass techniques, including .pyc bytecode bypass and charset smuggling, and covers 9 risk categories.
• Performs Jailbreak Evaluation with multi-turn attacks such as Many-Shot, PAIR, GOAT and ActorAttack.
• Ships a vulnerability library of 146 AI components and more than 2000 CVE rules for AI infrastructure scanning.
• Offers standalone skill-scan, mcp-scan and agent-scan CLIs plus an AI Security Skill Market with three official skills.
• Flags empty or incomplete MCP scans, preserves findings during context compaction and surfaces underlying connection errors.

Who uses it and how

• Security teams auditing an AI platform before release run the infrastructure scan against the 146-component and 2000-plus-CVE library.
• Agent developers validate skills before publishing, using the skill scanner's 9 risk categories and its reported SkillTrustBench score of 0.9848.
• MCP server maintainers run mcp-scan in dynamic mode to confirm that tool whitelisting blocks remote code execution paths.
• Red teams evaluating model deployment risk use the LLM API poisoning detection probes to test for model substitution and backdoors.
• CI pipelines and scripted workflows invoke skill-scan, mcp-scan or agent-scan directly, without the web frontend.

Getting started

Clone the repository and start the stack with Docker Compose using docker-compose -f docker-compose.images.yml up, which pulls pre-built images from Docker Hub. Requirements are Docker 20.10 or higher, 4GB or more RAM, and 10GB or more disk space.

How it compares

The facts for this entry name no comparable products and no paid alternatives that it replaces, so AI-Infra-Guard stands alone in this registry. Users evaluating it should treat the comparison as one against separate single-purpose scanners that would otherwise be combined by hand.

When to use it — and when not to

A self-hoster must operate a Docker host with at least 4GB of RAM and 10GB of disk, and no hosted or managed option appears in the project's quick start, so teams without a container platform should look elsewhere. Teams that do not run agents, MCP servers or skill packages will find little of the Agent Scan, Skill scan and ClawScan surface applies to them, leaving only the infrastructure scan and jailbreak evaluation. The project moves quickly, with releases through 2026, which means operators should expect frequent upgrades and a changelog to track.

# This method pulls pre-built images from Docker Hub for a faster start
git clone https://github.com/Tencent/AI-Infra-Guard.git
cd AI-Infra-Guard
# For Docker Compose V2+, replace 'docker-compose' with 'docker compose'
docker-compose -f docker-compose.images.yml up -d

clawhub install aig-scanner

# This method will automatically install Docker and launch A.I.G with one command
curl https://raw.githubusercontent.com/Tencent/AI-Infra-Guard/refs/heads/main/docker.sh | bash

git clone https://github.com/Tencent/AI-Infra-Guard.git
cd AI-Infra-Guard
# This method builds a Docker image from local source code and starts the service
# (For Docker Compose V2+, replace 'docker-compose' with 'docker compose')
docker-compose up -d