# MCP Tools and Usage

> The Sisense MCP Server exposes four read-only tools for discovering data sources, querying them in natural language, and building charts. Learn what each tool does and see example prompts.

*Source: https://docs.sisense.com/main/SisenseLinux/mcp-tools-usage.htm*

---

Last updated: August 15, 2026

|  |  |
| --- | --- |
| [Tier](https://www.sisense.com/pricing/#pricing) | [Deployment](https://docs.sisense.com/main/SisenseLinux/introduction-to-sisense-cloud-managed-services.md#ComparisonofManagedCloudandSelfHosted) |
| Launch     Grow    Enterprise | Cloud     On-Prem |

This page covers the tools the server exposes and how to use them. To connect a client first, see [Connecting an AI Agent](https://docs.sisense.com/main/SisenseLinux/mcp-connect-ai-agent.md).

## Available Tools

The server exposes four read-only tools. They only read data and metadata. They never modify your models, dashboards, or data.

### Discover

| Tool | What It Does |
| --- | --- |
| `getDataSources` | Lists the data models and data sources you can access, with their build status. This is how the agent finds what is available to query. |
| `getDataSourceFields` | Returns the tables, columns, relationships, and formulas of a chosen data source: the schema and semantic definitions the agent reads before it queries. Can be scoped to a specific perspective. |

### Query

| Tool | What It Does |
| --- | --- |
| `buildQuery` | Runs a natural-language query against a chosen data model and returns the result set. The query is resolved against your semantic model, so it uses your governed metrics and relationships. Results are cached in the session so later steps can build on them. When a result is too large to return in full, only a representative sample is shown, and the agent is prompted to build a chart with `buildChart` so you can explore the complete result visually. |

### Visualize

| Tool | What It Does |
| --- | --- |
| `buildChart` | Builds a chart from a previous query result, and can update its type, title, sorting, styling, and data options. Supported types include line (trends), bar and column (comparisons), pie (distributions), and area (cumulative); you can name the type in your request. Responses can include a short narrative summary of the result. In clients that support interactive content, the chart renders live in the conversation; other clients receive only the narrative insights, without a rendered chart. |

**Tip: Improving Results.**

Agents produce better answers when your data models are AI-ready: clear field and measure names, well-defined formulas, and curated perspectives. The more governed context your model carries, the more accurate the agent's queries and charts will be. See [Optimizing Your Data Model for AI](https://docs.sisense.com/main/SisenseLinux/ai-best-practices.md).

## Using the MCP Server

Once connected, you work with your data in plain language. A typical session moves from discovery to a specific chart:

1. **See what you can query.** Ask the agent to list your data sources, for example, "List my Sisense data sources."
2. **Inspect a model.** Ask for the fields of the source you want, so the agent knows the available dimensions and measures, for example, "What fields are in Sample ECommerce?"
3. **Ask a question or request a chart.** Describe what you want in business terms, for example, "Show total revenue by month for Sample ECommerce."
4. **Refine.** Build on the result: change the chart type, add a dimension, or narrow the range.

### Example Requests

**Discover**

- "List my Sisense data sources."
- "What fields are available in Sample ECommerce?"

**Ask and Visualize**

- "Show total revenue by month for Sample ECommerce."
- "Show the top 10 products by sales."
- "Compare sales by region."
- "Show revenue by category as a percentage of the total."
- "Show revenue comparison year over year."

### Tips for Good Results

- **Be specific.** "Show total revenue by month as a line chart, sorted chronologically" beats "show revenue."
- **Refine step by step.** Start simple, then add a trend line, a dimension, or a filter.
- **Use exact data source names.** Data source names are case-sensitive. If a request fails, ask for the list of sources and copy the exact name.
- **One data source per chart.** To combine sources, build separate charts.
- **Lean on common patterns:** time series ("... over time"), top-N ("top 10 ... by ..."), comparison ("compare ... by ..."), and distribution ("... distribution by ...").