Secure AI Infrastructure with cnspec
Scan AI platforms like Anthropic Claude and OpenAI, and AI infrastructure like Model Context Protocol (MCP) servers, Ollama instances, and vLLM inference servers, against security best practices with cnspec.
As AI moves into production, the platforms that serve your models and the servers that host models and tools become part of your attack surface. cnspec scans your AI platform accounts and AI infrastructure against security best practices so you can ship AI features without opening new holes.
New to cnspec? Read the Quickstart to install cnspec and run your first scan.
Choose what to scan
Anthropic Claude
Scan Claude organizations, workspaces, and API keys.
OpenAI
Scan OpenAI organizations, projects, and API keys.
MCP servers
Scan Model Context Protocol (MCP) servers for security misconfigurations.
Ollama
Scan Ollama instances for exposure and review the models they serve.
vLLM servers
Scan vLLM inference servers against security best practices.
Inventory your AI models
To see which AI models, agents, and services you run, generate an AI bill of materials (AIBOM) with cnspec aibom. It inventories models on your local system, in Ollama and Hugging Face, and in AWS (Amazon Bedrock and Amazon SageMaker), Google Cloud (Vertex AI and Model Armor), and Azure (Azure AI services):
cnspec aibom aws -o cyclonedx-jsonTo build an AIBOM for an Anthropic Claude organization, an OpenAI organization, or a vLLM server, define the asset and its credentials in an inventory file and pass it with --inventory-file. An inventory with several assets produces one AIBOM per asset. Output formats are Markdown (the default), JSON, CycloneDX JSON, and CycloneDX XML. To learn more, read the cnspec aibom reference.
Go further with Mondoo Platform
To continuously monitor your AI infrastructure and track posture over time, register cnspec with Mondoo Platform.