Windsor

Translates natural language questions into optimized queries against Windsor's integrated data warehouse without requiring users to write SQL. Uses LLM-driven query generation with schema awareness to map user intent to appropriate data sources, handling multi-table joins and aggregations transparently. The MCP protocol bridges the LLM's function-calling interface with Windsor's query execution engine, enabling conversational data exploration across connected business systems.

Provides automatic schema discovery and normalization across Windsor's integrated data sources (CRM, marketing platforms, analytics tools, databases, etc.), exposing a unified schema to the LLM through MCP's resource listing interface. The capability handles schema mapping, field aliasing, and relationship inference without manual configuration, allowing the LLM to understand available data without users manually documenting table structures or relationships.

Executes queries against Windsor's data warehouse and automatically contextualizes results with business metric interpretation, trend analysis, and anomaly detection. The capability combines query execution with post-processing logic that compares results against historical baselines, calculates growth rates, identifies outliers, and generates business-relevant insights without requiring users to manually specify analysis parameters or thresholds.

Enables LLMs to construct multi-step analytical workflows where intermediate query results are persisted and referenced in subsequent queries, supporting complex analysis patterns like cohort analysis, funnel analysis, and comparative metrics. The capability manages result caching and state across multiple MCP function calls, allowing the LLM to build sophisticated analyses without recomputing intermediate steps or losing context between queries.

Maintains synchronized copies of business data from source systems with configurable refresh intervals and freshness guarantees. The capability handles incremental syncs, change detection, and conflict resolution across multiple sources, exposing data freshness metadata to the LLM so it can make informed decisions about data reliability and whether to refresh before querying. Uses MCP's resource metadata to communicate sync status and last-update timestamps.

Enforces row-level and column-level access controls at query execution time, ensuring LLMs only access data they're authorized to query. The capability integrates with Windsor's permission model to filter query results based on user context, data classification, and compliance rules, preventing unauthorized data exposure while maintaining transparent access patterns that the LLM can understand and reason about.

Supports execution of multiple queries in batch mode with results exported to various formats (CSV, JSON, Parquet) and destinations (cloud storage, email, webhooks). The capability handles query queuing, parallel execution where possible, and result aggregation, enabling LLMs to request bulk data exports or schedule recurring analytical reports without blocking on individual query execution.

Implements intelligent query result caching with cache invalidation based on source data freshness, and exposes query execution plans to the LLM so it can understand performance characteristics and optimize queries. The capability tracks which source tables are referenced in each query, automatically invalidates cached results when those tables are updated, and provides execution time estimates to help the LLM make decisions about query complexity.