Back to MCP Servers

ThoughtSpot MCP Server

Remote, OAuth-authenticated MCP server that lets AI assistants query ThoughtSpot analytics data, generate answers, and build Liveboards.

Analytics by ThoughtSpot OAuth2 active
Overview

The ThoughtSpot MCP Server is an official, remote server hosted by ThoughtSpot on Cloudflare that exposes ThoughtSpot's analytics capabilities to LLM-based clients such as Claude, ChatGPT, and Gemini. Rather than self-hosting, users connect their MCP client directly to ThoughtSpot's hosted endpoint (https://agent.thoughtspot.app/mcp) and authenticate using OAuth (with Dynamic Client Registration) or a bearer token. The server supports both Server-Sent Events and streamed HTTP transports and offers date-based API versioning (including latest, dated versions, and beta).

The server's tools let an AI agent discover analytical questions, run natural-language data queries against ThoughtSpot data models, and persist findings as ThoughtSpot Liveboards. It exposes ping, getRelevantQuestions, getAnswer, createLiveboard, and getDataSourceSuggestions, plus a datasources MCP resource that lists data models the authenticated user can access. Newer API versions add Spotter 3 session-based conversation tools for multi-step reasoning.

Because the server is hosted by ThoughtSpot, setup is minimal: an active ThoughtSpot instance (or free trial), CORS and SAML whitelist entries for agent.thoughtspot.app, and either an OAuth login flow or a bearer token from the REST Playground. The open-source repo at github.com/thoughtspot/mcp-server is published by the official ThoughtSpot organization for transparency and contribution.

Tools

Tool Description
ping Verifies connectivity and authentication to the ThoughtSpot MCP endpoint.
getRelevantQuestions Generates a list of analytical questions relevant to the user's prompt against a given data model.
getAnswer Executes a natural-language analytical question against ThoughtSpot and returns the answer data.
createLiveboard Persists one or more answers as a ThoughtSpot Liveboard (dashboard).
getDataSourceSuggestions Recommends appropriate ThoughtSpot data models for a given query or topic.
datasources (resource) MCP resource listing ThoughtSpot data models accessible to the authenticated user.
Setup Guide

Prerequisites

  • An active ThoughtSpot instance (a free trial works)
  • Node.js 18+ on the machine running your MCP client (needed for the mcp-remote shim)
  • In ThoughtSpot Security Settings, whitelist agent.thoughtspot.app under both CORS whitelisted domains and SAML redirect domains

Option 1: OAuth (recommended for desktop clients)

Add this to your Claude Desktop (or other MCP client) config:

{
  "mcpServers": {
    "ThoughtSpot": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://agent.thoughtspot.app/mcp?api-version=latest"
      ]
    }
  }
}

On first use you will be redirected through a browser-based OAuth flow. The server uses Dynamic Client Registration, so no client ID/secret is needed.

Option 2: Bearer token (recommended for API / server-to-server)

  1. In ThoughtSpot, go to Develop → REST Playground v2.0 → Authentication → Get Full access token, or call https://<your-ts-instance>/api/rest/2.0/auth/session/token after logging in.
  2. Connect to the token endpoint with headers:
Authorization: Bearer $TS_AUTH_TOKEN
x-ts-host: your-instance.thoughtspot.cloud

Endpoints:

  • https://agent.thoughtspot.app/mcp (streamed HTTP, OAuth)
  • https://agent.thoughtspot.app/sse (SSE, OAuth)
  • https://agent.thoughtspot.app/token/mcp (streamed HTTP, bearer token)
  • https://agent.thoughtspot.app/token/sse (SSE, bearer token)

All endpoints accept an optional ?api-version=YYYY-MM-DD (or latest / beta) query parameter for version pinning.

Use Cases
  • Ask natural-language analytics questions ("What was revenue by region last quarter?") and get back ThoughtSpot answer data inside an LLM chat.
  • Have the agent suggest a relevant data model first, then iteratively refine analytical questions before running them.
  • Generate a set of relevant exploratory questions for a new dataset and let the user pick which to run.
  • Promote useful answers into a persistent ThoughtSpot Liveboard so business users can revisit them outside the chat session.
  • Embed ThoughtSpot analytics into agent workflows via the token endpoint for server-to-server automation.
Example Prompts
  • "List the ThoughtSpot data sources I have access to and recommend the best one for analyzing churn."
  • "Generate five relevant analytical questions for the Sales Performance data model."
  • "What were the top 10 products by revenue last month? Use the Retail Sales datasource."
  • "Take the last three answers and build a Liveboard called 'Q1 Exec Review'."
  • "Ping the ThoughtSpot MCP server to confirm my authentication is working."
Pros
  • Officially built and hosted by ThoughtSpot, so no self-hosting or infrastructure is required.
  • Supports both OAuth (with Dynamic Client Registration) and bearer-token auth, covering both interactive and programmatic use.
  • Goes beyond read-only queries: agents can persist findings as Liveboards inside ThoughtSpot.
  • Date-based API versioning (latest, dated, beta) lets clients pin behavior and opt into Spotter 3 session tools.
Limitations
  • Requires an active ThoughtSpot instance and admin access to add CORS/SAML whitelist entries for agent.thoughtspot.app.
  • Locked to ThoughtSpot's hosted endpoint; the repo is not designed as a drop-in self-hostable server.
  • Tool surface is focused on Q&A and Liveboards; lower-level admin, modeling, and data-loading operations are not exposed.
Alternatives