This glossary uses the current public Agent Skills and MCP documentation as its source base. Terms are intentionally concise so they can be cited, copied into planning docs, or used as a shared vocabulary when designing reusable agent capabilities.
Agent Skill
A reusable capability package for an AI agent, centered on a SKILL.md file with metadata and instructions, plus optional scripts, references, and assets.
SKILL.md
The required Markdown file in an Agent Skill. It contains YAML frontmatter such as name and description, followed by the instructions an agent reads when the skill activates.
Progressive disclosure
A staged loading pattern: the agent sees lightweight metadata first, then reads the full skill instructions when relevant, then loads supporting resources only as needed.
Tool
An executable function exposed to an AI application. Tools can query systems, call APIs, compute results, write files, or perform other actions.
MCP
Model Context Protocol, an open protocol for connecting AI applications to external systems through hosts, clients, and servers.
MCP host
The AI application that coordinates one or more MCP clients, such as an IDE, desktop assistant, or chat application.
MCP client
The connector inside an MCP host that maintains a connection to a particular MCP server.
MCP server
A program that exposes context and capabilities to MCP clients, commonly through tools, resources, and prompts.
MCP resource
URI-addressed contextual data exposed by an MCP server, such as files, database schemas, records, or application-specific information.
MCP prompt
A reusable interaction template or workflow exposed by an MCP server.
Skill description
The frontmatter field that tells an agent what a skill does and when to use it. It is central to discovery and activation.
Reference file
A supporting document bundled with a skill and loaded only when the active task requires more detail than SKILL.md should carry.
Asset
A bundled template, example, schema, image, or data file that supports a skill workflow.
Skill portability
The degree to which a skill can move across agent clients without rewriting its instructions, file paths, scripts, packaging, or runtime assumptions.