Sigma Computing MCP Server

The Sigma MCP server is an official, provider-hosted remote MCP endpoint that lets AI assistants interact with a Sigma Computing organization using natural language. It is delivered over HTTP transport with OAuth authentication, so the assistant inherits the calling user's existing Sigma permissions, including connection, workspace, and row/column-level security. Supported clients listed in the docs include Claude (via the Connectors Directory), Codex, Cortex Code, and Cursor.

The server is organized around three capability areas. Search lets agents locate documents and data sources (data models, warehouse tables, workbook elements) by name, type, owner, and connection, and browse collections of documents across the organization. Data exploration lets the agent inspect column names, definitions, data types, and other structural attributes of data sources and documents. Data analysis lets the agent query connections, data models, or workbooks in natural language without writing SQL, and ask follow-up questions to refine the analysis. Sigma has announced an additional Build capability for generating workbooks, dashboards, and data models from prompts.

Because the server is hosted by Sigma, there is no package to install or repo to clone. Users connect by retrieving a per-organization MCP URL from their Sigma profile and pointing their MCP client at it. Permissions and AI provider configuration are managed inside Sigma itself.

Tools

Prerequisites

• A Sigma Computing account with at least one of the following permissions: View connections, View data models, View workbooks.
• At minimum Can view access to target documents and Can use access to relevant connections.
• An AI provider must be enabled for your Sigma organization.

Get your Sigma MCP URL

In Sigma, open your profile and go to Profile > MCP > Sigma MCP. Copy the MCP URL listed there. This URL is specific to your organization.

Connect from your AI assistant

Claude: Add the connector from the Sigma Connectors Directory and complete the OAuth login.

Codex CLI:

codex mcp add sigma --url <Sigma MCP URL>

Cortex Code CLI:

cortex mcp add sigma <Sigma MCP URL> --transport http

Cursor: Edit mcp.json and add the server:

{
  "mcpServers": {
    "sigma": {
      "url": "<Sigma MCP URL>"
    }
  }
}

On first use, the client will trigger an OAuth flow against your Sigma account. No API keys or local installation are required.

• Ask "What data do we have on customer retention?" and have the agent surface relevant Sigma workbooks, data models, and warehouse tables.
• Inspect the schema of a data model or workbook element (column names, data types, descriptions) before writing analysis.
• Run ad hoc natural language questions against a governed connection or data model and iterate with follow-up prompts, all while respecting existing row, column, and workspace permissions.
• Locate a specific workbook by owner, type, or connection across a large Sigma organization without browsing the UI.
• Use Sigma's data and semantics from inside Claude, ChatGPT, Cursor, Codex, or Cortex Code as a single grounded analytics source for agentic workflows.

• "Search Sigma for any workbooks or data models related to monthly active users owned by the data team."
• "Describe the columns and data types in the Customer Retention data model."
• "Using the Sales connection, what was total revenue by region for the last 4 weeks?"
• "Find the workbook called Q1 Pipeline Review and list its pages and elements."
• "Drill into the previous answer and break revenue out by product line, then by week."

• Official, Sigma-hosted server with no install or self-management required.
• OAuth inherits existing Sigma permissions, including row/column-level security, so the assistant only sees what the user can see.
• Listed Claude connector and explicit support for Codex, Cortex Code, and Cursor over standard HTTP + OAuth.
• Natural language querying across connections, data models, and workbooks without writing SQL.

• Requires a Sigma organization with an AI provider enabled and appropriate user permissions; not usable without a Sigma account.
• Some SQL features are not supported in queries, including custom window frames, ordered window aggregates, and dynamic intervals.
• The full list of individual tool names is not published in the public docs; capabilities are described at the Search / Explore / Analyze level, and Build is announced but not yet generally available.

• Snowflake Cortex MCP / Cortex Analyst for natural language querying directly against Snowflake.
• Hex MCP server for AI access to Hex notebooks and data apps.
• Community Sigma REST API MCP wrappers (for example, listings on MCP Bundles) that expose granular admin operations like workspace, team, grant, and scheduled export management, at the cost of being unofficial.