# Sisense MCP Server (Beta) > The Sisense MCP Server lets any MCP-compatible AI agent explore your data and build charts through Sisense, under Sisense governance, secured with OAuth 2.1. *Source: https://docs.sisense.com/main/SisenseLinux/mcp-server.htm* --- # Sisense MCP Server (Beta) 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 | **Availability:** Managed (cloud) and self-hosted deployments · Sisense 2026.3.1 and later · Beta. - Enabled by default - Administrators control it under **System Configuration > MCP** - Full tool usage (the AI-powered tools `buildQuery` and `buildChart`) requires Cloud-Linked Features enabled on your instance The Sisense MCP Server lets any MCP-compatible AI agent, such as Claude, ChatGPT, or Cursor, explore your data and build charts through Sisense, under Sisense governance. It is a fully hosted endpoint secured with OAuth 2.1, so there is nothing to install and no credentials to share. Agents connect *through* Sisense rather than directly to your data. They work against your semantic model (your metrics, terms, and relationships), so the results reflect your own definitions, and every request runs as the signed-in user within that user's permissions. To connect an agent, see [Connect an AI Agent](https://docs.sisense.com/main/SisenseLinux/mcp-connect-ai-agent.md). For the tools and example prompts, see [MCP Tools and Usage](https://docs.sisense.com/main/SisenseLinux/mcp-tools-usage.md). ## What Is MCP? The Model Context Protocol (MCP) is an open standard that lets AI agents connect to external systems in a consistent way. An MCP client (the AI tool) connects to an MCP server (Sisense) and calls the tools the server exposes. You do not need to know the protocol to use this feature — your MCP client handles it. ## What You Can Do Once an agent is connected, it can: - **Discover your data.** List the data models and data sources you have access to, and inspect their fields, measures, and formulas before querying. - **Ask questions in natural language.** The agent describes what it wants in plain language, and Sisense translates it into a governed query against your model. - **Build and refine charts.** Turn a query result into a chart, then change its type, title, sorting, or styling. In agents that support interactive content (such as Claude), the chart renders live inside the conversation. - **Chain analysis across steps.** Results are kept in the session, so the agent can build on a previous query or chart instead of starting over. Because the agent works against your semantic model, it uses the same metrics and relationships your team already relies on, not ad-hoc definitions it invents on the fly. ## How It Works ``` AI agent --OAuth 2.1--> Sisense MCP Server --runs as you--> Your data model (Claude, (hosted endpoint) (metrics, terms, ChatGPT, ...) relationships) ``` 1. You add the Sisense MCP endpoint to your AI client as a server. 2. On first use, the client sends you through a standard OAuth 2.1 sign-in to Sisense, using your existing credentials, including SSO. 3. After you approve access, Sisense issues the agent a short-lived, per-user credential. The agent never sees your Sisense password or a shared key. 4. Every tool the agent calls runs as *you*, against *your* instance's data model, filtered by *your* permissions. There is no identity-provider configuration and no shared service account. Because the agent reaches your data only through Sisense, your data security, row and column permissions, and model definitions all apply automatically. ## Prerequisites - Your Sisense deployment is on **2026.3.1 or later**. - The MCP Server is available for your deployment. It is controlled by your Sisense administrator (**System Configuration > MCP**) and is on by default. - You have a Sisense user account with access to the data models you want to explore. The agent can only see what you can see. - You are using an MCP-compatible client that supports remote servers over HTTP and OAuth. **Claude**, **ChatGPT**, and **Cursor** are validated for the Beta. - For the AI-powered tools (`buildQuery` and `buildChart`), enable **Cloud-Linked Features** on your instance under **Admin > Sisense Intelligence > Feature Management**, then set up an LLM provider. You can use the LLM managed by Sisense, which draws on your Sisense credit package, or configure your own provider under **Provider Configuration**. The discovery tools (`getDataSources` and `getDataSourceFields`) read metadata only and do not require this. **Two Language Models, Two Roles.** The MCP Server relies on two distinct models. Your AI client (Claude, ChatGPT, or Cursor) supplies the model that drives the conversation and chooses which tools to call. Sisense uses its own instance-configured model to translate natural language requests into governed queries and charts against your data model. Cloud-Linked Features enables this Sisense-side model, which the AI-powered tools require. See the FAQ below for more. ## Security and Governance Access through the MCP Server is governed the same way as the rest of Sisense. - **No shared credentials.** Agents connect over OAuth 2.1. There is no shared API key or service account, and the agent never receives your Sisense password. Each connection uses PKCE and a short-lived, per-user credential that expires automatically. - **Every request runs as the signed-in user.** All tool calls execute under your identity and are filtered by your Sisense roles and data security, the same row- and column-level permissions you have in the product. Two people connecting to the same endpoint see different data, according to their own permissions. - **Attributable and logged.** Every request is tied to the signed-in user and recorded, so activity can be traced to a specific identity. **Where Your Data Goes:** When an agent queries through the MCP Server, the results are returned into the AI client you are using. For example, if you use ChatGPT, the query results are in ChatGPT. Treat the connected client as you would any tool that can see your data, and connect only clients you trust. ## Limitations (Beta) - **Read-only.** The tools explore data and build charts. They do not create or modify data models, dashboards, or saved content in Sisense. - **Sessions re-authenticate periodically.** Per-user credentials are short-lived, so long-running agents are prompted to sign in again when a credential expires. - **Interactive charts require a compatible client.** Clients that do not support MCP interactive content receive only narrative insights (a text summary of the chart), not a rendered chart. - **Auditing is log-based today.** Requests are logged and attributable to a user. A dedicated agent-activity admin view is planned for a later release. ## Frequently Asked Questions (FAQ) [**Q:** Do I need to install anything?](#) **A:** No. The endpoint is fully hosted. You add a URL to your AI client. There is no server to run, maintain, or update. [**Q:** Do I have to configure an identity provider?](#) **A:** No. The MCP Server reuses your existing Sisense login, including SSO. No IdP setup is required. [**Q:** Which LLM does the MCP Server use?](#) **A:** Two models work together. The model inside your AI client (for example, the one running in Claude or ChatGPT) manages the conversation and decides when to call a Sisense tool. Sisense then uses its own configured LLM, enabled through Cloud-Linked Features, to resolve natural-language requests into governed queries and charts against your data model. Because query and chart generation happen on the Sisense side, results stay consistent with your semantic model and permissions no matter which client model you use. This applies to the AI-powered tools (`buildQuery` and `buildChart`); the discovery tools read metadata and do not use an LLM. [**Q:** Can the agent change my data or dashboards?](#) **A:** No. All tools are read-only. They read data and metadata and build charts within the conversation. [**Q:** Will the agent see data I am not allowed to see?](#) **A:** No. Every request runs as you, filtered by your Sisense permissions. The agent can only reach what your account can reach. [**Q:** Which AI clients are supported?](#) **A:** Any MCP-compatible client that supports remote servers with OAuth. Claude, ChatGPT, and Cursor are validated for the Beta. [**Q:** Can I call the MCP Server from my own application?](#) **A:** Today you connect through an MCP client. Programmatic access for building your own agent or application against the server is planned for a future release. [**Q:** Is this the same as the local Sisense MCP server?](#) **A:** No. This hosted MCP Server replaces the earlier [local MCP server](https://docs.sisense.com/main/SisenseLinux/local-mcp.md), which ran on your machine and used a single shared token. The local server is being deprecated in favor of this hosted endpoint, which adds remote hosting and OAuth 2.1 so each user connects with their own identity and permissions. If you are still using the local server, we recommend moving to the hosted endpoint.