generated: '2026-08-27'
method: probed
source: https://help.sigmacomputing.com/docs/use-sigma-mcp-server + https://api.sigmacomputing.com/.well-known/oauth-protected-resource
  + POST https://api.sigmacomputing.com/mcp
name: Sigma MCP Server
status: published
description: Sigma ships a first-party remote MCP server that lets an AI assistant search the organization for workbooks,
  data models and warehouse tables, describe their columns and semantics, and run governed queries against them in natural
  language. Permissions are inherited from the caller's Sigma account; queries are tagged "kind":"mcp" in the warehouse so
  cost can be attributed.
deployment:
  mode: remote
  endpoint: https://api.sigmacomputing.com/mcp
  auth: oauth
  verified: probed
  note: 'The endpoint is per-organization: Sigma tells each customer to read their own Sigma MCP URL from Profile > Integrations
    > "Connect Sigma to AI tools". https://api.sigmacomputing.com/mcp is the GCP (US) cloud form and is the value Sigma itself
    advertises in /.well-known/oauth-protected-resource on that host; every regional API host serves the same shape (probed
    https://aws-api.sigmacomputing.com/mcp for AWS US-West). Transport HTTP, not stdio — there is no npx/pip package for this
    server.'
transport: http
protocol_version: unknown — server refuses initialize without a token
evidence:
- url: https://api.sigmacomputing.com/mcp
  method: POST tools/list
  http_status: 401
  body: '{"error":{"errorCode":"unauthorized","code":"EACCES","message":"Token missing or malformed"}}'
  www_authenticate: Bearer resource_metadata="https://api.sigmacomputing.com/mcp/.well-known/oauth-protected-resource", scope="mcp:access"
  note: A conformant RFC 9728 challenge. The live tool list and inputSchemas require an authenticated OAuth session, so the
    capability list below is transcribed from Sigma's own MCP documentation, not from tools/list.
- url: https://api.sigmacomputing.com/.well-known/oauth-authorization-server
  http_status: 200
- url: https://api.sigmacomputing.com/.well-known/oauth-protected-resource
  http_status: 200
authorization:
  scheme: OAuth 2.1 authorization code + PKCE (S256)
  issuer: https://api.sigmacomputing.com
  authorization_endpoint: https://api.sigmacomputing.com/mcp/authorize
  token_endpoint: https://api.sigmacomputing.com/mcp/token
  registration_endpoint: https://api.sigmacomputing.com/mcp/register
  dynamic_client_registration: true
  scopes:
  - mcp:access
  - api:access
  required_sigma_permission: Account type with "Use Sigma MCP with OAuth" enabled, plus the same view/use permissions the
    operation would need in the Sigma UI.
capabilities:
- name: search
  summary: Search the whole Sigma organization for documents and data sources — workbooks, data models, reports, warehouse
    tables, elements — filtered by name, type, owner or connection.
  source: docs
- name: describe
  summary: 'Explore a data source or document: column names, definitions, descriptions, data types, keys and column attributes.'
  source: docs
- name: query
  summary: Run a governed query against a connection, data model or workbook and return results, without the caller writing
    SQL. Follow-up refinement supported.
  source: docs
capabilities_note: Named capability groups as documented at https://help.sigmacomputing.com/docs/use-sigma-mcp-server. These
  are Sigma's feature headings, NOT verbatim MCP tool names — the real tool names and inputSchemas need an authenticated tools/list
  and have not been observed.
limitations:
- Custom window frames, ordered window aggregates and dynamic intervals are not supported in MCP-issued queries.
- Organizations with an IP allowlist must allowlist the AI assistant's outbound IPs with Scope=API.
distribution:
- client: Claude
  path: Connectors Directory
  url: https://claude.com/connectors/sigma
  status: 200
- client: ChatGPT
  path: ChatGPT plugin marketplace — search "Sigma"
- client: OpenAI Codex
  path: codex mcp add sigma --url <Sigma MCP URL>
- client: Snowflake Cortex Code
  path: cortex mcp add sigma <Sigma MCP URL> --transport http
- client: Cursor
  path: mcp.json custom server entry
other_mcp_servers:
- name: Sigma documentation MCP server
  endpoint: https://help.sigmacomputing.com/_mcp/server
  auth: none
  verified: probed
  server_info:
    name: fern-docs-mcp-server
    version: 1.0.0
    protocolVersion: '2025-06-18'
  tools:
  - name: searchDocs
    description: Search the documentation at https://help.sigmacomputing.com. Returns relevant doc passages with source URLs.
  note: Anonymous, live, and returns a real tools/list. This is the Fern docs-search server that ships with Sigma's documentation
    platform, advertised by Sigma in its own llms.txt — it reads documentation, it does not reach the Sigma product. Recorded
    separately from the product MCP server above so the two are not conflated.