[BUG] kusto_query takes 6–8 seconds for trivial queries due to per-request token acquisition #3865

Describe the bug

kusto_query takes approximately 6-8 seconds end-to-end for a trivial query whose Kusto-reported execution duration is under one second.

An additional independent run took 8.139 seconds. Every response succeeded and reported duration: 0:

{"status":200,"message":"Success","results":{"items":[{"value":"long"},[1]]},"duration":0}

Suggested investigation

Please measure these stages independently:

1. TokenCredential.GetTokenAsync
2. Kusto HTTP request execution
3. Kusto response parsing/transformation
4. MCP response serialization

Potential fixes include caching the Kusto access token by tenant/scope until shortly before expiration or using an Azure Core bearer-token authentication pipeline that handles token caching.

Expected behavior

For a trivial query with a known cluster URI, Azure MCP overhead should be small relative to direct Kusto REST execution.

Actual behavior

Timings

An additional independent run took 8.139 seconds. Every response succeeded and reported duration: 0:

{"status":200,"message":"Success","results":{"items":[{"value":"long"},[1]]},"duration":0}

Reproduction Steps

• Azure MCP Server: @azure/mcp@2.0.5
• Cluster URI: <doesn't matter>
• Database: <doesn't matter>
• Query: print value=1
• Transport: stdio
• Namespace: kusto

Environment

• OS: Windows
• Client: GitHub Copilot CLI 1.0.92
• MCP server: Azure MCP Server @azure/mcp@2.0.5
• Tool: kusto_query
• Transport: stdio
• Namespace: kusto
• Authentication: Azure credential chain / Azure CLI
• Runtime: Node.js v24.18.0
• Cluster: <any cluster url>
• Database: <any DB>