# Cube MCP server: 30 tools, remote (hosted), from Cube

SERVER, read 10 Nov 2026

**The short answer.** Cube MCP server is an official MCP server for Cube, published by Cube and read from its own source on 10 Nov 2026. It runs remote (hosted), and 30 tools are named. Version read: Hosted service (no versioned release). Last release: none found on 10 Nov 2026.

- **Data tool:** [Cube](/mcp/cube/)
- **Publisher:** Cube
- **Kind:** Official
- **Repository or docs:** [docs.cube.dev/docs/integrations/mcp-server](https://docs.cube.dev/docs/integrations/mcp-server)
- **Version or commit:** Hosted service (no versioned release)
- **Last release:** none found on 10 Nov 2026
- **Transport:** Remote (hosted)
- **Auth:** OAuth Authorization Code + PKCE (scope mcp-agent-access, public client cube-mcp-client), or a Cube API key exchanged for a token (client_id cube-api-key).
- **Access:** Read and write
- **Tools listed:** 30 named in the evidence
- **Tool definitions read at:** [docs.cube.dev/docs/integrations/mcp-server.md](https://docs.cube.dev/docs/integrations/mcp-server.md)
- **Context weight:** 1,093 tokens for 30 tool definitions (30 with a description), measured from the docs page read
- **Status:** Exists on 10 Nov 2026 as a Cube-hosted endpoint cubecloud.dev/mcp (or your own console domain). The page says it exposes 30 tools; the tables list 28 names, all captured here.
- **Read:** 10 Nov 2026

## In one sentence

Cube's hosted server searches the semantic layer, runs SQL with row-level security as the signed-in user and edits the semantic model on a branch.

## What Cube MCP server exposes

30 tools are named in the evidence read on 10 Nov 2026.

`listDeployments`, `chat`, `loadQueryResults`, `searchDataModel`, `runQuery`, `readWorkbook`, `createWorkbook`, `createReport`, `readReport`, `updateReport`, `deleteReport`, `manageFolders`, `updateDashboard`, `publishDashboard`, `listDataModelFiles`, `readDataModelFile`, `startDataModelEdit`, `writeDataModelFile`, `deleteDataModelFile`, `getDataModelChanges`, `getBranchDiff`, `getDeploymentEnv`, `getBranchState`, `commitDataModelChanges`, `commitToNewBranch`, `mergeToDefaultBranch`, `discardDataModelChanges`, `switchUserBranch`, `getPreAggregationStatus`, `buildPreAggregation`

## Not found in the sources read

- A lineage tool in the Cube MCP tool tables (getBranchDiff and getDataModelChanges compare model versions, not data lineage). [Source](https://docs.cube.dev/docs/integrations/mcp-server.md).

## The evidence on Cube MCP server, by topic

### Publisher and releases

- Cube hosts the MCP server and documents it on docs.cube.dev, available on all plans. (Cube's own claim; [source](https://docs.cube.dev/docs/integrations/mcp-server.md), read 10 Nov 2026.)
- No versioned release is published; one endpoint cubecloud.dev/mcp serves every account, deployment and region. (Cube's own claim; [source](https://docs.cube.dev/docs/integrations/mcp-server.md), read 10 Nov 2026.)

### Clients, auth and scoping

- Every tool runs as the authenticated user, and queries respect Cube permissions including row-level security. (Cube's own claim; [source](https://docs.cube.dev/docs/integrations/mcp-server.md), read 10 Nov 2026.)
- With an API key, a token can act as the key's owner, as another user, or be limited to specific tools. (Cube's own claim; [source](https://docs.cube.dev/docs/integrations/mcp-server.md), read 10 Nov 2026.)
- Which tools a user sees depends on their role; Viewer role or higher is needed, and data model editing tools show only for roles allowed to edit the model (Admin and Developer by default). (Cube's own claim; [source](https://docs.cube.dev/docs/integrations/mcp-server.md), read 10 Nov 2026.)

### Access and safety

- Nine tools are annotated destructive and clients that honor annotations always ask before running them. (Cube's own claim; [source](https://docs.cube.dev/docs/integrations/mcp-server.md), read 10 Nov 2026.)
- Chat is annotated read-only but runs the Analytics Chat agent, which can create reports or delete folders without confirmation, so the annotation understates what it can do. (Cube's own claim; [source](https://docs.cube.dev/docs/integrations/mcp-server.md), read 10 Nov 2026.)
- File writes never touch production: writeDataModelFile and deleteDataModelFile reject any branch that isn't a dev branch. (Cube's own claim; [source](https://docs.cube.dev/docs/integrations/mcp-server.md), read 10 Nov 2026.)
- API-key tokens last up to one hour and there is no refresh token. (Cube's own claim; [source](https://docs.cube.dev/docs/integrations/mcp-server.md), read 10 Nov 2026.)

### What it reaches

- Documented clients are Claude, ChatGPT, Microsoft 365 Copilot, Cursor and Codex, plus other MCP clients over HTTPS. (Cube's own claim; [source](https://docs.cube.dev/docs/integrations/mcp-server.md), read 10 Nov 2026.)
- Admins choose the default deployment and can limit which deployments MCP clients can reach, always intersected with role-based permissions. (Cube's own claim; [source](https://docs.cube.dev/docs/integrations/mcp-server.md), read 10 Nov 2026.)

### Lineage and impact

- SearchDataModel searches the semantic model by similarity, covering views and their measures and dimensions, and returns name, title, description and type. (Cube's own claim; [source](https://docs.cube.dev/docs/integrations/mcp-server.md), read 10 Nov 2026.)
- ListDataModelFiles and readDataModelFile read the raw YAML, JavaScript or Python source of the semantic model, so the definition logic itself is readable. (Cube's own claim; [source](https://docs.cube.dev/docs/integrations/mcp-server.md), read 10 Nov 2026.)

### Definitions, ownership and rules

- Model editing tools (startDataModelEdit, writeDataModelFile, commitDataModelChanges) let an agent change definitions on a personal dev branch, merged to the default branch only with a confirm-prompting tool. (Cube's own claim; [source](https://docs.cube.dev/docs/integrations/mcp-server.md), read 10 Nov 2026.)
- Query results carry the semantic layer: runQuery runs Cube SQL against views whose names and members come from searchDataModel. (Cube's own claim; [source](https://docs.cube.dev/docs/integrations/mcp-server.md), read 10 Nov 2026.)

### Freshness and upkeep

- The server can target a dev-mode branch, so agents can query a branch's model before it is merged. (Cube's own claim; [source](https://docs.cube.dev/docs/integrations/mcp-server.md), read 10 Nov 2026.)
- GetPreAggregationStatus and buildPreAggregation let an agent check and trigger pre-aggregation builds. (Cube's own claim; [source](https://docs.cube.dev/docs/integrations/mcp-server.md), read 10 Nov 2026.)

### Profile

- Admins need admin privileges or a custom role with the Manage MCP server and connectors permission to enable MCP. (Cube's own claim; [source](https://docs.cube.dev/docs/integrations/mcp-server.md), read 10 Nov 2026.)

## Read next

- [MCP servers for Cube](/mcp/cube/): 1 listed
- [Cube MCP server and dbt MCP Server, side by side](/compare/cube-mcp-vs-dbt-mcp/)
- [The Context Weight Index](/context-weight/)

## Sources

- docs.cube.dev/docs/integrations/mcp-server.md: https://docs.cube.dev/docs/integrations/mcp-server.md (read 10 Nov 2026)

Last reviewed 10 Nov 2026. Canonical: https://datamcpservers.com/servers/cube-mcp/
