IQ MCP Integration — Client...

No guardrails on data exposure

CRITICAL

Security

IQ/MCP integration allows natural language queries against all data without RLS enforcement. Sensitive HR and Finance data accessible to all users.

→ Implement MCP tool-level authorization checks. Enforce RLS through the semantic model connection. Add audit logging for all IQ queries.

3 semantic models missing descriptions

HIGH

Data Quality

Models without measure/table descriptions cause IQ to generate incorrect DAX 23% of the time. Only 'Finance Consolidated' has full descriptions.

→ Add descriptions to all tables, columns, and measures in Sales Revenue, Marketing Attribution, and Supply Chain models.

Ambiguous measure names causing wrong results

HIGH

Naming

Measures named 'Total' and 'Count' exist in multiple models. IQ cannot disambiguate without context, leading to wrong model selection.

→ Rename ambiguous measures with model prefix: 'Sales Total Revenue', 'HR Headcount Total'. Add synonyms for natural language mapping.

MCP tool timeout on large datasets

MEDIUM

Performance

3 of 12 MCP tools timeout when querying tables with >1M rows. Default 30s timeout is insufficient for complex aggregations.

→ Increase MCP tool timeout to 60s. Add pre-aggregated summary tables for common IQ queries. Cache frequent results.

Missing error handling in generated tools

MEDIUM

User Experience

MCP tools return raw API errors to end users when queries fail. No friendly error messages or fallback behavior.

→ Add try/catch wrappers with user-friendly error messages. Implement fallback to suggest alternative queries.

No usage analytics configured

LOW

Monitoring

No tracking of which IQ queries are asked, which succeed/fail, or which models are most queried.

→ Enable Application Insights integration. Log query text, model used, success/failure, and response time.

Findings (6)