If the AI returns empty results, missing values, or data that doesn’t match expectations when querying Cube through MCP, the issue is usually related to how the AI interpreted your prompt and which dimension members it selected.
This article explains how Cube MCP fetch logic works, how to inspect the query the AI generated, and how to troubleshoot missing or unexpected data.
How Cube MCP Fetching Works
The Cube MCP Server uses the same fetch logic as the Cube Spreadsheet Add-on.
When the AI sends a query to Cube, it builds a range using specific dimension members—just like a spreadsheet fetch. If your prompt does not specify a dimension member, the AI falls back to the Spreadsheet Add-on default behavior:
- It selects the parent dimension member with the largest number of child members.
This default behavior is expected, but it may not match the level of detail you intended.
As a result, you may see:
- Rolled-up data instead of detailed child-level values
- No data if the selected parent has no direct values
- Unexpected results if the parent includes accounts or entities outside your intended scope
The more specific your prompt, the more accurate the AI’s fetch will be. Include account names, scenarios, departments, entities, and time periods whenever possible.
Step 1: Understand Default Dimension Behavior
When prompts are broad—such as:
“Show me revenue this quarter”
—the AI does not know which specific dimension members to use unless you provide them.
In those cases, Cube defaults to a high-level parent member.
This can cause three common outcomes:
| Outcome | Explanation |
|---|---|
| Data appears too high-level | The AI queried a parent member instead of a detailed child member |
| No data is returned | The selected parent has no direct values for the requested period or scenario |
| Data looks incorrect | The parent includes additional accounts or entities you did not intend to query |
Best Practice
Use explicit dimension member names whenever possible.
Instead of:
“Show me revenue this quarter”
Try:
“Show me Total Revenue from the Actuals scenario for Q1 2025 by department.”
Step 2: Inspect the AI’s Tool Use or Thinking Section
Every MCP query includes the exact parameters the AI sent to Cube.
Reviewing this section is the fastest way to diagnose missing or unexpected data.
Open the Tool Use Section
- Locate the AI response that returned incorrect or empty data.
- Expand the section labeled:
- Thinking
- Tool use
- Used Cube
- or similar wording depending on the AI client
- Review the Cube query details.
What to Look For
Pay close attention to:
| Parameter | What to Verify |
|---|---|
| Dimension IDs | Confirm the correct members were queried |
| Scenario | Verify the correct scenario was used (Actuals, Budget, Forecast, etc.) |
| Time period | Confirm the requested month, quarter, or year matches your expectation |
| Filters | Check whether filters excluded the data unintentionally |
In Claude, this section typically appears as a collapsible “used Cube” block above the AI response.
Step 3: Match Dimension IDs to Your Cube Workspace
Once you identify the dimension IDs used in the query, confirm which members they represent in Cube.
Find a Dimension Member by ID
- Log in to your Cube Workspace.
- Navigate to Dimensions.
- Search for the dimension member or browse the hierarchy.
- Open the member and compare its ID to the ID shown in the AI’s tool use section.
You can also paste the ID directly into the Workspace URL.
Example format. The dimension ID is the last set of numbers in the URL:
Ex. https://portal.cubesoftware.com/fpa/dimensions/73496/edit/111389Validate the Member
Once located, confirm:
- The member is the expected account, department, entity, or scenario
- The member is at the correct hierarchy level
- The member contains data for the requested time period
- The member is mapped correctly to source data
Step 4: Refine Your Prompt
The most effective way to improve MCP query accuracy is to provide more precise prompts.
Use exact dimension member names from Cube whenever possible.
Example Prompts
| Broad Prompt | Improved Prompt |
|---|---|
| “Show me revenue this quarter.” | “Show me Total Revenue from the Actuals scenario for Q1 2025 by department.” |
| “What is our headcount cost?” | “Show me Salaries & Wages from the FY2025 Budget scenario for all departments.” |
| “How are we tracking against budget?” | “Compare Actuals vs. Budget for Operating Expenses by department for the last 3 months.” |
Prompting Best Practices
Include:
- Scenario names
- Actuals
- Budget
- Forecast
- Explicit time periods
- January 2025
- Q3 FY2024
- Last 12 months
- Exact account names
- Total Revenue
- Operating Expenses
- Salaries & Wages
- Desired breakdowns
- By department
- By entity
- By product line
If you’re unsure which dimension members exist, ask the AI:
“What dimension members are available for Accounts in Cube?”
The AI can query Cube metadata and return available members.
Step 5: Verify the Underlying Data
If the AI still returns no data after refining the prompt, the issue may be related to the underlying Workspace configuration rather than the MCP query itself.
Validate the Following
Source Data Imports
Go to:
Workspace → Integrations
Confirm:
- Data has been imported successfully
- The expected time period exists
- The correct scenario contains data
Dimension Mapping
Verify:
- Accounts are mapped correctly
- The queried dimension member is not unmapped
Unmapped members always return empty results.
Time Rollup Settings
Check whether the dimension member uses:
- No time rollup
- Restricted aggregation behavior
These settings can prevent quarterly or yearly totals from appearing.
Clearing Accounts
If clearing accounts are configured, data may be landing there instead of the intended member.
Test by fetching the clearing account directly.
Spreadsheet Add-on Validation
Build the same range manually in the Spreadsheet Add-on.
If the Spreadsheet Add-on also returns no data:
- The issue is with the data or configuration
- Not the MCP Server
The Spreadsheet Add-on Validate feature is a useful cross-check when troubleshooting MCP queries.
Quick Diagnostic Checklist
Use this checklist when the AI returns empty or unexpected results.
| Check | Action |
|---|---|
| Inspect tool use | Review the dimension IDs and filters the AI sent to Cube |
| Match dimension IDs | Confirm the correct dimension members were queried |
| Verify hierarchy level | Ensure the AI queried the intended child-level member |
| Confirm scenario | Validate the correct scenario was used |
| Confirm time period | Verify the requested dates align with available data |
| Refine the prompt | Use exact dimension member names |
| Validate in Spreadsheet Add-on | Recreate the same fetch manually |
| Check source imports | Confirm data exists for the requested period |
Still Need Help?
If the issue persists after completing the steps above, contact Cube Support and include:
- Your original prompt
- A screenshot or copy of the AI’s tool use/thinking section
- The dimension members you expected to query
- Whether the same fetch works in the Spreadsheet Add-on
Providing this information upfront helps Cube Support diagnose the issue more quickly.