Cube
Type: low-code · Vendor: Cube Dev, Inc. · Language: TypeScript, Rust · License: Apache-2.0 · Status: active · Status in practice: mature · First released: 2019
Define metrics and dimensions once in a governed semantic layer and serve them to AI agents and BI tools over SQL, REST, GraphQL, and an MCP server.
Description. Cube is a semantic layer that defines business metrics, dimensions, joins, and access rules once and exposes them through multiple query interfaces. Its open-source core, Cube Core, is released under Apache 2.0. AI agents reach the governed metrics over SQL, REST, GraphQL, and an MCP server rather than authoring raw SQL against the warehouse. MCP-compatible clients such as Claude and Cursor connect to Cube to get analytics answers grounded in the same model.
Agent loop shape. An AI agent connects to Cube through its MCP server (or another supported interface), poses a natural-language analytics question, and Cube resolves it against governed metrics and dimensions defined in the semantic layer, returning an answer grounded in the shared model rather than executing free-authored SQL on raw tables.
Primary use cases
- serving governed metrics to AI agents over an MCP server
- defining business metrics once for reuse across tools
- querying a semantic layer from BI and analytics interfaces
Key concepts
- Semantic layer (data model) → semantic-layer-query-guardrail (docs) — The single definition of business metrics, dimensions, joins, and access rules that every downstream interface and agent queries against, so the same numbers come back regardless of the tool.
- Governed metrics → semantic-layer-query-guardrail (docs) — Metrics defined once in the model and served as vetted, reusable measures, so an agent selects and parameterises an approved metric rather than computing it ad hoc from raw tables.
- Cube MCP server → mcp (docs) — The Cube connector that any MCP-compatible client adds to ask natural-language analytics questions and get answers grounded in the live, governed model.
- Row-level, multi-tenant access control → tenant-scoped-tool-binding (docs) — Access rules defined in the model that restrict which rows and metrics each user or tenant can reach, enforced uniformly across every query interface including the agent's.
Patterns this low-code implements —
- ★Semantic-Layer Query Guardrail
Cube routes AI-agent data questions through a governed semantic layer so agents query vetted metrics and dimensions over SQL/REST/GraphQL/MCP rather than free-authoring raw SQL against the warehouse.
- ★★Model Context Protocol
Cube exposes its governed metrics through an MCP server so MCP-compatible agents like Claude and Cursor connect and query the semantic layer through the standard protocol.
- ★Tenant-Scoped Tool Binding
Access control is defined once in the semantic layer and enforced beneath the agent: Cube applies row-level, multi-tenant access control so a connected agent or user only ever sees the data they are…
Neighbourhood
Click any neighbour to follow the lineage. Scroll to zoom, drag to pan.
Alternatives & relatives
Listed as alternative by (1)
References
Provenance
- Last analyzed:
- Last updated:
- Verification status: partial