Reference
Glossary
The words this course uses, in plain language.
- Agent
- An AI that takes actions (edits files, runs commands, calls services), not just one that answers.
- MCP
- Model Context Protocol. An open standard for connecting AI agents to tools.
- MCP server
- A program that exposes tools to an agent over MCP. Usually runs on your machine.
- MCP client
- The app that runs the agent and connects to servers: Claude Code, Cursor, VS Code and others.
- Tool description
- Text a server provides so the agent knows when to use a tool. The agent trusts it.
- Prompt injection
- Text that makes a model follow instructions its user didn't give.
- Tool poisoning
- Prompt injection hidden in a tool's description.
- Typosquatting
- Publishing a package whose name is one typo away from a popular one.
- Supply chain
- Everything your software depends on, and everyone who could change it.
- CVE
- A public ID for a known vulnerability, like CVE-2026-30615.
- OSV
- An open vulnerability database and format. mcp-audit's advisory feed uses it.
- Toxic flow
- Two servers whose combined abilities create an attack path.
- Attack path
- A chain of capabilities from your data to an attacker.
- Rug pull
- A server that changes after you approved it.
- Exfiltration
- Getting data out of a system without permission.
- Lock file
mcp-lock.json: a committed record of approved servers that CI can verify.- SARIF
- A standard format for scanner results that GitHub's Security tab reads.
- SBOM
- Software bill of materials: a list of every component in a piece of software.
- Severity
- How serious a finding is: CRITICAL, HIGH, MEDIUM, LOW, INFO.
- OWASP MCP Top 10
- OWASP's list of the ten main MCP risk categories, MCP01–MCP10.
- Shadow server
- An MCP server running somewhere without approval.
- Offline
- Runs without network calls. mcp-audit's default.