# Connecting an AI Agent > Connect an MCP-compatible AI client, such as Claude, ChatGPT, or Cursor, to the Sisense MCP Server by adding the hosted endpoint and signing in with your Sisense credentials. *Source: https://docs.sisense.com/main/SisenseLinux/mcp-connect-ai-agent.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 | Note: This page covers connecting an AI client. For what the MCP Server is and how it is governed, see [Sisense MCP Server](https://docs.sisense.com/main/SisenseLinux/mcp-server.md). You connect an MCP-compatible AI client to Sisense by adding the hosted endpoint and signing in once with your Sisense credentials. There is nothing to install. **Claude**, **ChatGPT**, and **Cursor** are validated for the Beta. Validation covers connecting and running the tools; whether a client can display a *rendered* chart depends on MCP Apps support in that client, described under [Interactive Charts in Claude](#InteractiveChartsInClaude) below. **Note for Administrators:** The MCP Server is enabled by default under **System Configuration > MCP**, on managed and self-hosted instances, with no extra deployment setup. To stop AI clients from connecting to a given environment, turn the **Enabled** toggle off there. Each Sisense environment is controlled independently. ## 1. Get the Endpoint URL The hosted MCP Server runs at a single Sisense address. There are two ways to tell it which deployment to work against, and both are supported. **Option 1: Name your deployment in the URL.** Add a `target` query parameter set to your Sisense URL: ``` https://mcp.sisense.com/mcp?target= ``` The client then connects straight to your deployment with no further prompting. This is the form administrators usually distribute when rolling the connector out to a team, because it settles both the endpoint and the deployment up front and nobody has to know or type their instance URL. **Option 2: Use the base endpoint and choose your deployment on screen.** Add the endpoint without a `target`: ``` https://mcp.sisense.com/mcp ``` The first time you connect, Sisense presents a form where you enter your deployment URL before signing in. This suits individual users, and anyone who connects to more than one deployment. Your Sisense administrator can confirm the exact endpoint and your deployment URL. On self-hosted deployments, the endpoint is exposed by your own instance and serves that instance directly, so no `target` parameter is needed: ``` https:///mcp ``` Your administrator provides the exact endpoint. ## 2. Add It to Your MCP Client Add the endpoint as a **remote (HTTP) MCP server**. There are no credentials to enter. Authentication happens in the browser on first use. [Claude (Desktop / claude.ai)](#) 1. Open **Settings > Connectors > Add custom connector**. 2. Give it a name (for example, `Sisense`) and paste your endpoint URL. 3. **Save**. Claude prompts you to sign in the first time you use it. To see interactive charts rendered inside Claude, your administrator must allow Claude's content domain in Sisense CORS settings. See [Interactive Charts in Claude](#InteractiveChartsInClaude) below. [ChatGPT](#) 1. Open **Settings > Connectors**. 2. Add a custom connector, name it (for example, `Sisense`), and paste your endpoint URL. 3. **Save**, then sign in to Sisense when ChatGPT prompts you. Connector availability depends on your ChatGPT plan. [Cursor](#) Add the server to your MCP configuration: ``` { "mcpServers": { "sisense": { "url": "https:///mcp?target=" } } } ``` Reload the MCP servers. Cursor opens a browser sign-in the first time the server is used. [Other MCP Clients](#) Any client that supports remote MCP servers over HTTP with OAuth can connect. Add the endpoint as a remote/HTTP server (not a local command). The client discovers Sisense's OAuth configuration automatically and registers itself. No manual client ID or secret is required. ## 3. Sign In and Approve Access The first time the agent uses the server, your browser opens a Sisense sign-in page. Sign in with your existing Sisense credentials, including SSO, and approve access. Sisense issues the agent a short-lived credential scoped to you, and you are ready to go. Note: **Each user connects individually.** Every person who wants to use the MCP Server completes this sign-in with their own account. There is no shared login, and each user sees only the data their permissions allow. ## 4. Verify the Connection Confirm the agent is connected before you rely on it: 1. Check that the Sisense tools appear in your client's tool or connector list. 2. Ask the agent to list your data sources, for example, "List my Sisense data sources." 3. You should see the data models your account can access. If the tools do not appear or the request fails, see [Troubleshooting](#Troubleshooting) below. ## Switching Deployment or Instance Each connection is scoped to one deployment and one user. - To point an agent at a **different Sisense deployment**, change the `target` in the endpoint URL. The client prompts you to sign in again for that deployment. - **Signing in as a different user**, or connecting to a different instance, also starts a new sign-in. Credentials are never shared between users or deployments. ## Interactive Charts in Claude Claude can render the charts built by `buildChart` as live, interactive visualizations inside the conversation. For this to work, your administrator must add Claude's content domain to the allowed origins in **Admin > Security Settings > CORS Allowed Origins**: ``` https://*.claudemcpcontent.com ``` Use the subdomain wildcard form. Claude loads each chart from an Anthropic subdomain (for example, `https://abc123.claudemcpcontent.com`), so the browser origin is a subdomain, not the bare domain. Listing `https://claudemcpcontent.com` on its own does not match, and the chart stays blocked. Note: **Rendered Charts Depend on MCP Apps, Not on MCP Compatibility Alone.** Sisense renders charts through MCP Apps, an extension for interactive content inside the conversation that not every MCP client implements. A client can be fully MCP compatible, connect successfully, and run every tool, and still have no way to display a rendered chart. Claude is the client validated for interactive charts in the Beta. In other clients, expect the narrative insights instead, and treat chart rendering as unverified until you have tried it in that client. Without this entry, queries and other tools still work, but the interactive chart will not load. Clients that do not support interactive content receive only the narrative insights from the chart, not a rendered chart. ## Troubleshooting [**The Sisense tools do not appear in my client.**](#) Reload or restart the MCP client, and confirm it supports remote MCP servers over HTTP. Check that the endpoint is entered as a remote/HTTP server, not a local command. [**I am asked to sign in again, or a request fails with an authentication error.**](#) Per-user credentials are short-lived. When one expires, the client prompts you to sign in again; complete the sign-in and retry. If it persists, confirm you are signing in to the correct Sisense instance. [**A request says the data source cannot be found.**](#) Data source names are case-sensitive. Ask the agent to list your data sources and use the exact name. [**The tools work, but no chart appears.**](#) Check whether your client supports MCP Apps. Clients without it receive the narrative insights rather than a rendered chart, and nothing is wrong with the connection or the query. In Claude, if the chart area stays blank, see the CORS entry under [Interactive Charts in Claude](#InteractiveChartsInClaude) above. [**The interactive chart does not render in Claude.**](#) Your administrator must allow Claude's content domain in Sisense CORS settings. See [Interactive Charts in Claude](#InteractiveChartsInClaude) above. Other tools still work without it. [**The chart is not what I expected.**](#) Be more specific: name the chart type, the fields, and any sorting or filter. Simplify the request, then build it back up. `buildQuery`**or**`buildChart`**fails, but discovery still works.** The AI-powered tools need Cloud-Linked Features and a configured LLM provider. Ask your administrator to confirm Cloud-Linked Features is enabled under **Admin > Sisense Intelligence > Feature Management**, and that an LLM provider is set up: either the LLM managed by Sisense, which draws on your Sisense credit package, or your own provider under **Provider Configuration**. If you use the Sisense-managed LLM, also check that your Sisense credit package has not run out. The discovery tools (`getDataSources` and `getDataSourceFields`) do not need this and keep working.