Aleph
Query live FP&A data, build reports and update dashboards from your AI assistant.
Cube
Query Cube semantic models, build dashboards and edit data models from an AI agent.
Opens Agent Studio, where connecting is one click. The connector URL below works in any MCP client.
Paste it into any MCP client. Setup docs
What the Cube connector does
The Cube MCP server connects a Cube semantic layer to Claude and any MCP-compatible agent. Its 23 tools run queries against governed data models, author dashboards and reports, edit the data model itself, and manage pre-aggregations. Model edits are confined to a personal dev branch by design.
Connection verified
Connection checked by Agentman on .
Listed in Anthropic’s Connectors Directory at community tier — a listing, not a partner-level review.
Anthropic states this reflects the level of review a connector received, not a security audit.
Cube tools (23)
- buildPreAggregation
- chat
- createReport
- createWorkbook
- deleteDataModelFile
- deleteReport
- getBranchDiff
- getDataModelChanges
- getDeploymentEnv
- getPreAggregationStatus
- listDataModelFiles
- listDeployments
- loadQueryResults
- publishDashboard
- readDataModelFile
- readReport
- readWorkbook
- runQuery
- searchDataModel
- startDataModelEdit
- updateDashboard
- updateReport
- writeDataModelFile
Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.
Limits
- Premium and Enterprise plans only, Viewer role or higher.
- Tool visibility varies by role, so what an agent can do differs per user on the same deployment.
- No commit tool — an agent cannot publish model changes; a person commits from the Cube UI.
valid: trueis a compile check, not a correctness check. Verify on the dev branch withrunQuery.- Admin configuration is a prerequisite. If Admin → MCP Server reports MCP configuration unavailable, the host is not set up for the account.
- Reports and dashboards are not branch-protected the way the model is —
deleteReportacts directly. - We did not read tool schemas. The endpoint requires OAuth, so names come from Anthropic's directory and behaviour from Cube's documentation. We exercised no tool.
Frequently asked questions
Which Cube plan do I need for the MCP server?
Premium or Enterprise. Cube states the MCP server is available on those plans only, and that users need the Viewer role or higher to interact with it. Which tools a user actually sees depends on their role, so two people on the same deployment can be offered different capabilities from the same endpoint.
Can an AI agent break my production Cube data model?
No, by construction. Cube documents that every model write goes to a personal dev branch named dev-user-hash, and that the write tools reject any branch that is not a dev branch, so the deploy branch is never writable over MCP. Promotion to production requires a person committing the branch in the Cube UI.
How does Cube stop an agent publishing model changes?
It exposes no commit tool at all. Cube states this is deliberate: an AI client can prepare changes on a dev branch, but only a person can ship them, by committing from the Cube UI. Combined with dev-branch-only writes, that keeps the promotion step human without relying on a client approval prompt.
Do I need one Cube endpoint per deployment or region?
No. Cube documents cubecloud.dev/mcp as the same URL for every account, deployment and region, identifying your account from the OAuth token and routing each request to whichever deployment serves it. Region-specific endpoints also exist under Admin then MCP Server, but the single URL is the documented default.
What does valid true mean from Cube writeDataModelFile?
Only that the model compiles. Cube is explicit that it does not mean the edit is correct, and recommends passing the dev branchName to runQuery to check the edit returns the right data. Verification happens on the branch before a merge rather than after, which is the point of the dev-branch workflow.
Can a Cube Viewer edit the data model through MCP?
No. Cube documents that tool registration is permission-gated: the six model-editing tools, getBranchDiff, getDeploymentEnv and both pre-aggregation tools are offered only to users whose role allows editing the semantic model. A Viewer never sees them at all, so the restriction is on tool visibility rather than on call failure.
Sources
- Cube MCP server documentation (read as Markdown via the
.mdvariant; retrieved 2026-09-13).docs.cube.dev/robots.txtcarriesContent-Signal: ai-train=yes, search=yes, ai-input=yes— an explicit publisher opt-in to AI synthesis, recorded here because we rely on it. · retrieved 2026-09-13 - Live OAuth posture check against
https://cubecloud.dev/mcp— HTTP 401 with awww-authenticateBearer challenge, and RFC 9728 metadata athttps://cubecloud.dev/.well-known/oauth-protected-resource/mcpreturning the single scopemcp-agent-access(2026-09-13) · retrieved 2026-09-13 - Anthropic Connectors Directory entry (retrieved 2026-09-13) · retrieved 2026-09-13
- Cube support — <support@cube.dev> · Privacy
More Analytics connectors
august
Interview simulated customer twins, run focus groups and surveys from an AI agent.
ChartMogul
Query MRR, ARR, churn, retention, cohorts, customers and sales pipeline from your AI assistant.
Product names and logos belong to their owners. Agentman is not affiliated with a connector's maker unless the page says the connector is built and run by Agentman.
Use it in an agent
Put Cube to work.
Connect Cube once and your agents call these tools on their own: on a schedule, in a workflow, with nobody at the keyboard.