The Problem: Natural Language Queries for Business Users
Business users want to ask data questions in plain English, not learn SQL or KQL. While Claude can generate queries, manually copying them into database tools is inefficient. The solution: connect Claude directly to your Azure Data Explorer (Kusto) database using a Remote MCP Server.
What MCP Servers Actually Do
Model Context Protocol (MCP) servers bridge Claude with external systems. For database access, you have two options:
- Local MCP Server: Runs on your machine, uses local credentials
- Remote MCP Server: HTTP-based, ready for production with proper authentication
This follows Anthropic's November 2024 introduction of MCP as an open standard for connecting AI systems to external tools and data sources.
Architectural Decision: Where MCP Fits in Microservices
The author presents a crucial insight: MCP shouldn't be a separate microservice. Instead, it should co-exist with REST endpoints in your existing query-handling microservices. This aligns with the CQRS+ pattern where MCP becomes just another protocol for exposing the same business logic.
In practice, if you're building a Remote MCP Server for Kusto querying, host it alongside your existing query-handling microservice within the Reporting bounded context.
Building Your Own Kusto MCP Server
While Microsoft may eventually offer an official Remote MCP Server for Kusto (they're reportedly working on one for Azure AI Foundry), you can build your own today. The protocol is JSON RPC over HTTP, similar to building a REST API.
Key components:
- Authentication: Handle oAuth for secure access
- Query translation: Convert natural language to KQL (Claude helps here)
- Result formatting: Return data in Claude-friendly formats
- Visualization support: Leverage Claude Desktop's split-screen view for charts
The author used .NET/F# with Claude Code writing 75% of the implementation, referencing Microsoft's Fabric RTI MCP Server as inspiration.
Why This Matters for Claude Code Users
As a Claude Code user, you're already comfortable with MCP servers for development tasks. Extending this pattern to database access opens new possibilities:
- Data exploration during development: Query production-like data without leaving your workflow
- Debugging with real data: Test assumptions against actual datasets
- Performance analysis: Check query patterns and optimization opportunities
Getting Started
If you work with Azure Data Explorer, consider building a simple MCP server:
# Example: Starting a basic MCP server project
claude code "Create a simple MCP server in TypeScript that connects to Kusto"
Focus on:
- Secure authentication (Azure AD)
- Schema discovery (list tables/columns)
- Query execution with proper error handling
- Result formatting for Claude's visualization capabilities
This approach gives you natural language query capabilities today, without waiting for vendor solutions.
