@sassoftware/sas-score-mcp-serverjs

What it does

This MCP server connects Claude to SAS Viya, enabling scoring operations—executing SAS models, jobs, and custom programs—through natural language. Users can invoke pre-trained models from Model Studio or Intelligent Decisioning, run SAS Studio flows, or execute user-written SAS programs directly from chat. The server also provides tools to query and list tables within SAS analytics environments. Supports both HTTP and stdio transport protocols with flexible authentication (OAuth, bearer token, CLI-based, and password flows).

Who it's for

SAS administrators and data scientists who operate SAS Viya and want to expose analytics workflows to Claude without switching contexts. Teams building AI-assisted decision support systems that rely on existing SAS models or custom programs as backend logic.

Common use cases

• Score a dataset against a pre-trained SAS model by passing table names and model identifiers via chat
• Execute a SAS Studio flow or job definition with parameters specified in natural language
• Query SAS table metadata and row counts from CAS or compute contexts
• Fetch model details and decision logic from SAS Model Studio for debugging and explanation
• Integrate SAS decisioning workflows into Claude-driven business processes and decision support

Setup pitfalls

• OAuth (the default auth method) requires the server to be deployed remotely (Azure, container, etc.) for Claude—localhost deployments cannot complete the OAuth callback
• VIYA_SERVER is required and must point to a valid SAS Viya instance; misconfigured URLs will fail silently during auth negotiation
• CAS and compute sessions are not cached per request, so repeated operations create new sessions and incur longer latency than expected
• OAuth redirect URI must exactly match your deployment URL; mismatches prevent token exchange and leave the connection unauthenticated