Connects Claude to your ThoughtSpot instance for natural language data queries. Offers OAuth or bearer token authentication via a hosted Cloudflare worker. Exposes tools for searching data sources, generating analytical answers with inline charts, and creating Liveboards from query results. Supports versioned endpoints with the api-version parameter, letting you pin to specific releases or track latest. Works across OpenAI's responses API, Claude's MCP connector, and Gemini SDK. Ships with streamable HTTP and SSE transports for real-time streaming. Useful when you want to ask conversational questions about your analytics data and turn answers directly into dashboards without leaving your AI workflow. The demo shows selecting a datasource as context, asking for top products, then generating a multi-chart Liveboard from the results.
The ThoughtSpot MCP Server provides secure OAuth-based authentication and a set of tools for querying and retrieving relevant data from your ThoughtSpot instance. It's a remote server hosted on Cloudflare.
If you do not have a Thoughtspot account, create one for free here.
Learn more about ThoughtSpot.
Join our Discord to get support.
See our privacy statement here.
As of May 1, 2026, the ThoughtSpot MCP Server supports Spotter 3, enabling advanced analytics, forecasting, multi-step reasoning, and deep research. The new MCP tools support real-time streaming and session-based conversations.
The server uses date-based versioning via the ?api-version=YYYY-MM-DD query parameter. Appending this to your MCP server URL pins your integration to a specific version (or the closest earlier version if an exact match doesn't exist). Each new version may introduce new tools, enhancements, or bug fixes.
OAuth apps (plug-and-play integrations for Claude, ChatGPT, or custom OAuth apps):
?api-version=2026-04-30.Bearer token apps (custom apps using bearer tokens):
?api-version=2026-04-30.To configure this MCP server in your MCP client (such as Claude Desktop, Windsurf, Cursor, etc.) which do not support remote MCPs, add the following configuration to your MCP client settings:
{
"mcpServers": {
"ThoughtSpot": {
"command": "npx",
"args": [
"mcp-remote",
"https://agent.thoughtspot.app/mcp?api-version=latest"
]
}
}
}
See the Troubleshooting section for any errors / more details.
See the video below for a complete demo.
Here is a demo video using Claude Desktop.
https://github.com/user-attachments/assets/72a5383a-7b2a-4987-857a-b6218d7eea22
Watch on Loom
The following examples show a single conversation that builds progressively, starting from data discovery through to a finished Liveboard.
User prompt: "What data sources are available for analyzing sales performance?"
What happens:
User prompt: "What are my top selling products this year?"
What happens:
User prompt: "Create a Liveboard for me from the findings"
What happens:
ThoughtSpot's remote MCP server can be used in LLM APIs which support calling MCP tools.
Here are examples with the common LLM providers:
curl https://api.openai.com/v1/responses \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $OPENAI_API_KEY" \
-d '{
"model": "gpt-4.1",
"tools": [
{
"type": "mcp",
"server_label": "thoughtspot",
"server_url": "https://agent.thoughtspot.app/token/mcp",
"headers": {
"Authorization": "Bearer $TS_AUTH_TOKEN",
"x-ts-host": "my-thoughtspot-instance.thoughtspot.cloud"
}
}
],
"input": "How can I increase my sales ?"
}'
Note: Use /token/mcp for new integrations. The older /bearer/mcp endpoint is deprecated but still supported for backward compatibility.
More details on how can you use OpenAI API with MCP tool calling can be found here.
curl https://api.anthropic.com/v1/messages \
-H "Content-Type: application/json" \
-H "X-API-Key: $ANTHROPIC_API_KEY" \
-H "anthropic-version: 2023-06-01" \
-H "anthropic-beta: mcp-client-2025-04-04" \
-d '{
"model": "claude-sonnet-4-20250514",
"max_tokens": 1000,
"messages": [{
"role": "user",
"content": "How do I increase my sales ?"
}],
"mcp_servers": [
{
"type": "url",
"url": "https://agent.thoughtspot.app/token/mcp",
"name": "thoughtspot",
"authorization_token": "$TS_AUTH_TOKEN@my-thoughtspot-instance.thoughtspot.cloud"
}
]
}'
Note: In the authorization_token field we have suffixed the ThoughtSpot instance host as well with the @ symbol to the TS_AUTH_TOKEN. Use /token/mcp for new integrations. The older /bearer/mcp endpoint is deprecated but still supported for backward compatibility.
More details on Claude MCP connector here.
MCP tools can be used with the Gemini Python/Typescript SDK. Here is an example using typescript:
import { GoogleGenAI, FunctionCallingConfigMode , mcpToTool} from '@google/genai';
import { Client } from "@modelcontextprotocol/sdk/client/index.js";
import { StreamableHTTPClientTransport } from "@modelcontextprotocol/sdk/client/streamableHttp.js";
// Create server parameters for stdio connection
const serverParams = new StreamableHTTPClientTransport(new URL("https://agent.thoughtspot.app/token/mcp"), {
requestInit: {
headers: {
"Authorization": "Bearer $TS_AUTH_TOKEN", // Read below how to get the $TS_AUTH_TOKEN
"x-ts-host": "my-thoughtspot-instance.thoughtspot.cloud"
},
}
});
const client = new Client(
{
name: "example-client",
version: "1.0.0"
}
);
// Configure the client
const ai = new GoogleGenAI({});
// Initialize the connection between client and server
await client.connect(serverParams);
// Send request to the model with MCP tools
const response = await ai.models.generateContent({
model: "gemini-2.5-flash",
contents: `What is the weather in London in ${new Date().toLocaleDateString()}?`,
config: {
tools: [mcpToTool(client)], // uses the session, will automatically call the tool
// Uncomment if you **don't** want the sdk to automatically call the tool
// automaticFunctionCalling: {
// disable: true,
// },
},
});
console.log(response.text)
// Close the connection
await client.close();
Note: Use /token/mcp for new integrations. The older /bearer/mcp endpoint is deprecated but still supported for backward compatibility.
Read this, for more details on Gemini API MCP tool calling.
An example using Google ADK + Python can be found here.
ThoughtSpot MCP server can also be installed as a Gemini CLI extension.
gemini extensions install https://github.com/thoughtspot/mcp-server
Read more about Gemini CLI here.
For API usage, you would the token endpoints with a secret_key to generate the API_TOKEN for a specific user/role, more details here.
ping: Test connectivity and authentication.getRelevantQuestions: Get relevant data questions from ThoughtSpot analytics based on a user query.getAnswer: Get the answer to a specific question from ThoughtSpot analytics.createLiveboard: Create a liveboard from a list of answers.getDataSourceSuggestions: Get datasource suggestions for a given query.datasources: List of ThoughtSpot Data models the user has access to.For MCP hosts which do not(yet) support Dynamic client registration, or they require statically adding Oauth Client Id etc. Go to this page, to register a new client and copy the details over. The most relevant values are Oauth Client Id and Oauth Client Secret which should be added when adding ThoughtSpot as an MCP connector in the MCP client (ChatGPT/Claude etc). The generated client details are only available when they are generated and are NOT available later for reference.
Manual client registration also allows to associate with a specific ThoughtSpot instance, so that your users do not have to enter the Thoughtspot instance URL when doing the authorization flow. While registering the Oauth client add ThoughtSpot URL to the appropriate value.
TS_AUTH_TOKEN ?TS_AUTH_TOKENOauth errors due to CORS/SAML.
Make sure to add the following entries in your ThoughtSpot instance:
CORS
SAML (need to be Admin)
MCP server install error due to node issues
node -v500 error from MCP server
Stale MCP auth
rm -rf ~/.mcp-auth.npm install
.dev.vars and fill in your ThoughtSpot instance URL and access token.npm run dev
When adding new MCP tools to the server:
src/servers/tool-definitions.tssrc/servers/mcp-server.tssrc/servers/version-registry.ts:
VERSION_REGISTRYDEFAULT_VERSIONbeta version entryImportant: The version registry controls which tools are available in each API version. Make sure to add new tools to the correct version configuration to ensure they're accessible to users.
OAuth-based endpoints:
/mcp: MCP HTTP Streaming endpoint (supports ?api-version)/sse: Server-sent events for MCP (supports ?api-version)/authorize, /token, /register: OAuth endpointsToken-based endpoints (Recommended for APIs):
/token/mcp: MCP HTTP Streaming with bearer auth (supports ?api-version)/token/sse: Server-sent events with bearer auth (supports ?api-version)Deprecated endpoints:
/bearer/mcp, /bearer/sse: Legacy MCP endpoints with bearer auth (deprecated, no api-version support). Use /token/* endpoints instead.API Versioning:
The ThoughtSpot MCP Server supports API versioning to access different tool sets. You can specify the version using the api-version query parameter on OAuth and /token/* endpoints (not supported on deprecated /bearer/* endpoints).
Version Formats:
?api-version=beta - Access the latest beta features?api-version=YYYY-MM-DD - Access tools from a specific release date or the latest version on or before that dateExamples:
# Beta version (latest experimental features)
https://agent.thoughtspot.app/token/mcp?api-version=beta
# Specific date version (Spotter 3 agent tools)
https://agent.thoughtspot.app/token/mcp?api-version=2026-05-01
# Date range resolution (returns newest version ≤ specified date)
https://agent.thoughtspot.app/token/mcp?api-version=2025-03-15
Available Versions:
beta: Latest beta features with Spotter3 agent conversation toolslatest: Most recent non-beta version2026-05-01: Spotter 3 agent conversation tools (create_analysis_session, send_session_message, get_session_updates)2025-01-01: Base MCP tools (getRelevantQuestions, getAnswer, getDataSourceSuggestions)Note: The /bearer/* endpoints always return the default stable tool set and ignore the api-version parameter for backward compatibility.
MCP Server, © ThoughtSpot, Inc. 2025
com.exploit-intel/eip-mcp
dmontgomery40/pentest-mcp
pantheon-security/notebooklm-mcp-secure
cyanheads/pentest-mcp-server
io.github.akhilucky/ai-firewall-mcp