mcp

This ThoughtSpot MCP Server provides secure OAuth-based authentication and a set of remote MCP tools to query and retrieve data from ThoughtSpot instances. It’s designed to be consumed by clients that support MCP tooling (such as Claude, OpenAI with tools, Gemini SDKs, and other MCP-enabled platforms). The server exposes a collection of tools, including ping, getRelevantQuestions, and getAnswer, which enable connectivity checks, contextual question retrieval from ThoughtSpot analytics, and direct question answering against ThoughtSpot data. To use the MCP server, configure your MCP client to point at the server URL and authenticate with a TS_AUTH_TOKEN or equivalent token as required by your integration. In clients that don’t natively support remote MCPs, you can embed the server as an MCP endpoint via the provided configuration example, which uses npx to launch the remote MCP tool wrapper.

Prerequisites:

• Node.js and npm installed on your machine (for using npx to run the remote MCP wrapper).
• Access to a ThoughtSpot instance and a valid authentication token.

Step-by-step:

1. Install Node.js and npm if you don’t have them:

# On macOS with Homebrew
brew install node

# On Windows or Linux, download from https://nodejs.org/

2. Install and run the remote MCP wrapper using npx (as configured in mcp_config):

npx mcp-remote https://agent.thoughtspot.app/mcp

3. If you need to register the server in a MCP client that does not support remote MCPs, use the following JSON configuration in your client settings:

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

4. Obtain a TS_AUTH_TOKEN (as described in the docs) and configure your API clients to pass the token in the Authorization header when calling MCP tools.

Notes:

• The server is hosted as a remote MCP endpoint; ensure network access to https://agent.thoughtspot.app/mcp is allowed.
• If you’re integrating with an MCP client that requires a different startup flow, consult the client’s documentation for tool invocation using the server URL and headers.

Tips and common considerations:

• Ensure your ThoughtSpot instance is accessible from the environment running the MCP client (consider firewall rules and network egress).
• Tools exposed by the MCP server include ping, getRelevantQuestions, and getAnswer; use them to validate connectivity and fetch contextual analytical data.
• When using API-based integrations (OpenAI, Claude, Gemini), pass the TS_AUTH_TOKEN and the target ThoughtSpot host via the headers as shown in the examples.
• If you encounter authentication errors, verify that the TS_AUTH_TOKEN is valid for the user/role and that the token has not expired.
• For local development, you can test the MCP flow by invoking the wrapper with curl or your preferred HTTP client against the remote MCP URL once authenticated.