ThoughtSpot MCP

ThoughtSpot's MCP server exposes its analytics engine to AI clients so an agent can ask a business question and receive an answer computed from the customer's governed data model rather than from a stale extract. The analytics tools accept natural language queries, resolve them against the semantic model, and return results. Beyond returning numbers, the server can create Liveboards and visualizations programmatically, and generate embeddable versions of them, so a conversation that starts as a question can end with a shareable board rather than a screenshot. Tool discovery is automatic, meaning a connected client picks up the available ThoughtSpot tools without manual registration. The governance story is the reason this differs from pointing a model at a warehouse connection. Queries execute inside ThoughtSpot, so row level and column level security rules apply to every request, and an agent returns only the rows and columns the authenticated user is entitled to see. That makes it viable to expose analytics to a broad internal audience through a chat client. Two endpoints are documented: one for OAuth applications and one for bearer token applications, with the server authenticating the request and routing it to the ThoughtSpot instance. It works with MCP compatible platforms including Claude, ChatGPT and Gemini clients. Availability is tied to Enterprise Edition or Embedded licensing, and the intended users are organizations that want to add conversational analytics to an existing product or internal workflow without building a semantic layer and query engine themselves.