XLTable: OLAP server for Excel and AI assistants v 2.1.2

XLTable is an OLAP server that gives your analytical warehouse two familiar front ends: Excel PivotTables and an AI assistant. To Excel it looks like Microsoft Analysis Services: users build ordinary PivotTables — hierarchies, drill-down, slicers, filters — and XLTable turns every action into SQL that runs inside your warehouse. For AI assistants (Claude, Cursor, Yandex AI Studio, any MCP client) it ships a built-in MCP server: the assistant asks questions against the same cubes, under the same permissions, and gets the same figures the user sees in Excel.

How this differs from “connecting AI to the database”

• The assistant does not write SQL against raw tables. It works with a cube — the measures, dimensions and metrics the data owner defined once. XLTable builds the SQL.
• The engine does the arithmetic, not the model: period comparison, top-N and shares, trend and outliers, variance from plan, change decomposition by driver (“92% of the drop is the Clothing category”). The model understands the question; the server computes the numbers.
• One set of permissions for everyone: roles, field visibility and row-level filters apply in Excel and in the chat alike. The assistant physically cannot show rows the user is not entitled to.
• The assistant sees what the user is looking at in Excel: the layout of the open PivotTable can be continued with “explain this figure” — no need to restate the context.

What business users get

• Excel PivotTables with no SQL, add-ins or extracts: drill-through to rows, hierarchies, slicers, Keep Only / Hide, full-table CSV export.
• Questions in plain language: “why did sales fall in September”, “which stores give 80% of profit”, “how does actual deviate from plan by region”.
• A semantic layer for AI: cube and field descriptions, synonyms, live sample values, cube author’s instructions — the assistant knows that “revenue” and “turnover” are the same thing and filters by “North” rather than guessing “North Region”.

What IT gets

• Data never leaves the warehouse: no copies, no extracts, no intermediate databases — only queries to your DBMS from a read-only account. Shared result cache, parallel workers, overload guard.
• Cubes are plain SQL files: Git, pull requests, CI/CD. A cube is generated from a table automatically; an AI agent writes a cube definition from DDL.
• Access: local users, Active Directory / LDAP, Kerberos single sign-on, OAuth 2.1 for AI clients, roles and row-level filters defined in the cube.
• Air-gapped option: works with local models (LM Studio, Jan) — warehouse, cubes and assistant stay inside your perimeter.
• An Analysis Services replacement when you move to ClickHouse or Snowflake: the same PivotTables, hierarchies and MDX, with the data staying where it lives.

What is in the image
Ubuntu 24.04 LTS, XLTable 2.1.2 (server edition), Apache on ports 80/443, workers under supervisor, Yandex Cloud root certificates for connecting to Managed Service for ClickHouse. 2 vCPU / 4 GB is enough for a pilot; plan 4+ vCPU / 16 GB for production load.

Administrator console and documentation are in English.

1. Create a virtual machine from the image. Open ports 80 and 443 (and 22 for SSH) in the security group.
2. Open the administrator console at http://<IP>/admin. Default users are user1 / pass1 and user2 / pass2 (file /usr/olap/xltable/setting/settings.json); change the passwords before going live.
3. On the License page copy the Server ID and send it to help@xltable.com or Telegram t.me/XLTable — a trial license is issued free of charge; upload the .lic file on the same page.
4. Set the warehouse connection (host, port, user) in settings.json — for Managed ClickHouse this is port 8443 with TLS, the certificates are already in the image — and restart the service: sudo supervisorctl restart 'olap:*'.
5. Create a cube definition: a SQL file as described in the documentation (Cubes section) or the olap_definition table in the warehouse; for the first cube, take the ready-to-run sample for your DBMS from the documentation.
6. Excel: Data → Get Data → From Database → From Analysis Services, server http://<IP>, user name and password of an XLTable user. AI assistant: MCP server address https://<IP>/mcp, sign-in by OAuth 2.1 (same user’s login and password).
7. Optional: your own TLS certificate and Active Directory single sign-on — re-run the installer /usr/olap/install_xltable.sh with --cert/--key and --auth ad (see the Installation section of the docs).

Connecting to Yandex AI Studio. In the MCP tool settings enter the address https://<domain>/mcp (an Internet-reachable address with a trusted certificate is required) and Bearer authorization with the value base64(login:password) of an XLTable user. The assistant gets the tools list_cubes, describe_cube, query_cube and the analytical operators with that user’s permissions.

• Finance and controlling: plan vs. actual and margin in Excel PivotTables directly on ClickHouse, with no extracts and no BI layer in between.
• “Why did margin fall in September?” — the assistant decomposes the change by category and store through the cube and names the main driver.
• A regional manager sees only their own stores — in Excel and in the chat with the assistant: one set of permissions.
• Migration from Microsoft Analysis Services to ClickHouse or Snowflake, keeping the PivotTables of hundreds of users.
• An analyst connects a new table: the cube is generated automatically, the PivotTable and the assistant’s answers are minutes away.
• Air-gapped: a local model in LM Studio plus XLTable inside the perimeter — not a single query leaves for the cloud.

CloudReports

The XLTable team provides technical support to XLTable users: OLAP server for ClickHouse in Yandex Cloud.

Contact us:

• Telegram: t.me/XLTable
• Email: help@xltable.com

Yandex Cloud
Yandex Cloud does not provide support for this product. If you have any issues, please refer to the vendor’s information resources.