Sisense MCP Server

A Model Context Protocol (MCP) server that provides integration with Sisense analytics platform. This server enables LLMs to interact with Sisense data models and create charts programmatically.

Features

• Transport: Streamable HTTP (streamable-http) for HTTP-based MCP clients (for example Claude Desktop, Cursor)
• MCP tools (three by default; optional fourth when enabled):
  • getDataSources: Retrieve Sisense data sources (or data models)
  • getDataSourceFields: List all available fields for a specific data source
  • buildChart: Build charts from natural language prompts
  • buildQuery (optional): Run analytics queries when TOOL_BUILD_QUERY_ENABLED / toolBuildQueryEnabled is enabled
• MCP Apps: When used in MCP Apps–capable clients (for example Claude), buildChart exposes an interactive View that renders the chart in an iframe within the app.
• Per-session authentication: Sisense credentials via URL parameters and/or server environment variables
• TypeScript: Full type safety and modern ESM support
• Lightweight: Pure Node.js HTTP server, no heavy frameworks
• Fast: Optimized for Bun runtime, also runs on Node.js

Documentation

• Quick start — clone, .env, run, MCP client setup
• Configuration — credentials, tunneling, feature flags, URL examples
• FAQ — common questions and troubleshooting
• Usage examples — prompts and workflows

Prerequisites

• Node.js >= 18.0.0 (required for local development and npm)
• Bun for running project scripts (dev, build, start, tests): the bun package is a devDependency, so after npm install or bun install the Bun binary is available under node_modules/.bin—a global Bun install is not required. A global Bun install is optional.
• Sisense instance with API access
• Sisense API token
• Cloud-Linked Features enabled on your Sisense instance (and an LLM provider configured if using Bring Your Own Key (BYOK)). Required for natural-language tools such as buildChart and buildQuery. See Generative AI (Cloud-Linked Features). This is separate from your MCP client's LLM (e.g. Claude in Cursor).
• Playwright Chromium (installed automatically by bun install / npm install via postinstall)

Installation

bun install
# or
npm install

Usage

Start the server:

# Development mode (hot reload)
bun run dev
# or
npm run dev

# Production mode
bun run build && bun run start
# or
npm run build && npm run start

Sessions are in-memory — chart state is lost if the server restarts.

The server prints something like the following (port defaults to 3001, or PORT if set):

Sisense MCP Server running on http://localhost:3001

Connect with:
  http://localhost:3001/mcp?sisenseUrl=<SISENSE_URL>&sisenseToken=<SISENSE_TOKEN>
  Or set SISENSE_URL and SISENSE_TOKEN in the environment and use http://localhost:3001/mcp

Optional feature-flag query params (override env vars per connection):
  mcpAppEnabled=true|false, toolBuildQueryEnabled=true|false, toolBuildChartNarrativeEnabled=true|false

Endpoints:
  Health: http://localhost:3001/health
  Screenshots: http://localhost:3001/screenshots/

Connecting your MCP client

Use an MCP streamable HTTP URL. For Cursor, Claude Desktop, and similar clients, add a server entry with the MCP path (not a shell command such as bun run dev or npm run dev).

If SISENSE_URL and SISENSE_TOKEN are set in the server environment (for example in .env loaded by the process that runs bun run dev or npm run dev), the client URL does not need to include credentials:

{
  "mcpServers": {
    "sisense-analytics": {
      "url": "http://localhost:3001/mcp"
    }
  }
}

Note: Depending on your network or client environment, the localhost HTTP setup may not connect. In those cases, you will need to expose your local server publicly via HTTPS using a proxy service such as ngrok. Point the client at your HTTPS tunnel URL with the same /mcp path (and query parameters if you are not using server env credentials).

Credentials: If you do not use server env vars, put sisenseUrl and sisenseToken on the MCP URL as query parameters (URL params take precedence over env when both are present). Always percent-encode each value — see Configuration: URL encoding for details and examples.

Alternative connection patterns (placeholders only; use encoded values for real credentials):

http://localhost:3001/mcp?sisenseUrl=https://your-instance.sisense.com&sisenseToken=your-api-token

With SISENSE_URL and SISENSE_TOKEN in the server environment only:

http://localhost:3001/mcp

Behind a public HTTPS tunnel (example):

https://your-ngrok-url.ngrok-free.app/mcp?sisenseUrl=https://your-instance.sisense.com&sisenseToken=your-api-token

Claude (claude.ai and Desktop): Sisense CORS for interactive charts

In MCP App mode (the default), Anthropic Claude renders charts inside the MCP content UI—on claude.ai or Claude Desktop. The browser loads your Sisense instance from origins under *.claudemcpcontent.com, so Sisense must allow those origins or the widget fails with a network or CORS error.

In your Sisense instance, go to Admin → Security Settings → CORS Allowed Origins and add:

https://*.claudemcpcontent.com

Use this subdomain wildcard form. A single origin such as https://claudemcpcontent.com (no *.) may not match the actual frame origins and charts can still fail to load.

More context: FAQ: Claude MCP App charts and CORS.

Configuration

The server automatically derives its public base URL from request headers, so it works correctly behind proxies like ngrok. For how to build encoded MCP URLs, see URL encoding for query parameters.

Optional feature-flag query parameters

Defaults suit most setups; change flags when you need a specific client behavior. For when to use each flag, copy-paste URL patterns, and env vs query string, see docs/guides/configuration.md.

These query params override the corresponding env vars on a per-connection basis. Accepted values: true, false, 1, 0 (case-insensitive).

Example URL with all three overrides (encode sisenseUrl and sisenseToken values when they are not simple alphanumeric placeholders):

http://localhost:3001/mcp?sisenseUrl=https://your-instance.sisense.com&sisenseToken=your-api-token&mcpAppEnabled=false&toolBuildQueryEnabled=true&toolBuildChartNarrativeEnabled=false

Development

# Run server in development mode with hot reload
bun run dev
# or npm run dev

# Build the project (View + server)
bun run build
# or npm run build

# Build only the analytics View (dist/view.html)
bun run build:view
# or npm run build:view

# Run tests
bun test
# or npm test (same as npm run test)

# Type checking
bun run type-check
# or npm run type-check

# Lint
bun run lint
# or npm run lint

Security Considerations

⚠️ NEVER commit credentials to version control

⚠️ Use secret managers or vaults - NOT environment variables in production

⚠️ NEVER bind to 0.0.0.0 in production - use 127.0.0.1 or Unix socket

⚠️ Recommended: Use dev or staging Sisense when you have them. Autonomous AI clients can issue many API calls, and prompts can be ambiguous. Non-production reduces the impact of mistakes and surprises.

⚠️ Enable authentication - never run without auth

⚠️ Approve EVERY tool call - review all parameters before execution

⚠️ Create dedicated Sisense service account with minimum required permissions

⚠️ Rotate credentials regularly (every 90 days recommended)