Agent discovery
Finding every AI agent and MCP server running across an organization, along with who runs it and what it can access. Discovery is the first step in securing agents.
Glossary
AI agent security glossary, in plain words.
This AI agent security glossary defines the terms you'll run into when securing AI agents and MCP servers.
AI agent
Software that uses a language model to plan and take actions on its own, such as editing code, running commands, or calling other services. Coding assistants like Claude Code and Cursor are AI agents.
AI bill of materials (AI-BOM)
An inventory of the AI models, agents, and tools a system depends on. It does for AI what a software bill of materials does for code libraries.
AI security posture management (AI-SPM)
The practice of continuously finding, scoring, and fixing the security risks in an organization’s AI agents, models, and tools.
Data exfiltration
Moving data out of an organization without permission. For agents this often means a tool reading secrets like an .env file and sending them to an outside host.
MCP server
A program that exposes tools to an AI agent over the Model Context Protocol. Examples include servers that read files, query a database, or post to a chat app. Each one widens what the agent can reach.
Model Context Protocol (MCP)
An open standard that lets AI agents connect to outside tools and data through a common interface. It is how most agents gain new abilities today.
NoFly list
NoFly’s public reputation list of MCP servers and agent tools. Each entry has a risk score explaining whether it is safe for an agent to load.
Prompt injection
An attack that hides instructions inside content an agent reads, such as a web page, file, or tool response, so the agent follows the attacker instead of the user.
Shadow AI
AI agents and tools used inside a company without the security team knowing or approving them. Shadow AI is the agent version of shadow IT.
Tool poisoning
A prompt injection placed in an MCP tool description or response. The agent trusts the tool, so the hidden instructions can make it leak data or run commands.