MCP Glossary

The table below lists some common terms encountered when working with MCP.

Term	Definition
Graphmart	A logical container for graph data in Graph Studio. Graphmarts organize data into layers and can be deployed to Graph Lakehouse for querying.
Layer	A subdivision of a graphmart . Source layers load raw data; transform layers apply mappings.
Step	An individual processing unit within a layer, such as a SPARQL query, file load, or transformation.
Graph Lakehouse	The graph query engine that powers graphmart queries. Graphmarts must be activated to a Graph Lakehouse server before they can be queried.
Ontology	An RDF schema that defines the classes, properties, and relationships in your data model.
Skill	A named group of related MCP tools with shared context, enabling progressive disclosure to AI assistants. Skills can also contain sub-skills for hierarchical organization.
Sub-Skill	A child skill referenced by a parent skill. When the AI resolves the parent, it sees the sub-skill listed and can drill in further. A sub-skill marked as "sub-skill only" is hidden from the top-level catalog.
Shared Context	Instructions delivered once when a skill is expanded, replacing duplicated per-tool guidance
MCP Tool	A function exposed to AI assistants through the Model Context Protocol.
Skills Mode	Server configuration controlling which tools are directly visible vs. discoverable through the skills catalog.
Dynamic Skill	An auto-generated skill bucket that groups uncataloged tools by their name prefix (e.g., service_* → "Other Semantic Services"). Only appears in SKILLS_ONLY and ALL_TOOLS modes.