Part 2: Stop Asking Around. Start Asking Claude.

In the first installment of our four-part series introducing CloudQuant Data Liberator, we illustrated how companies that struggle to access and manage large data sets can save time, money, and resources by using Liberator's zero-ETL architecture to provide instant, secure access to new datasets,  ready for AI queries.

In this post, we’re talking about talking. Specifically, talking to Claude, or your AI Agent of choice. 

The Traditional Data Discovery Problem

Picture this scenario:

A new analyst joins your team. They need to find customer purchase data. They ask around:

• "Check the sales database"
• "Actually, customer data is in Salesforce"
• "But transactions are in the payments DB"
• "And reconciliation reports are in that shared drive somewhere..."

Three days later, they're still hunting for the right data. And they haven't even started analyzing it yet.

Enter the Model Context Protocol

We built an MCP (Model Context Protocol) server for Data Liberator. This lets AI agents like Claude directly query your data while respecting all your security, entitlements, and access controls.

Here's what that conversation looks like now:

Analyst: "Show me all datasets I have access to."

Claude: You have access to 47 datasets. Here are the ones related to customers and purchases:

• customer_profiles - Customer demographic and account data, updated daily
• transaction_history - All purchase transactions, real-time
• payment_reconciliation - Daily reconciliation from finance team

Analyst: "What's in the transaction_history dataset?"

Claude: The transaction_history dataset contains:

• customer_id - Links to customer_profiles
• transaction_date - When purchase occurred
• amount - Purchase amount in USD
• product_category - What they bought

Analyst: "Show me total purchases by category for the last month."

Claude: [Executes query, returns formatted results with a chart]

Three days of hunting became three minutes of conversation.

This Isn't a Chatbot

This is fundamentally different from hard-coded chatbots or SQL generators. Claude:

• Discovers what datasets you have access to
• Reads the SME-authored descriptions to understand context
• Understands your business terminology from dataset metadata
• Generates appropriate queries based on your question
• Respects your entitlements—only shows data you're allowed to see

There are no pre-programmed queries. Claude figures out what you need and how to get it.

How We Built It: Enterprise-Grade Security

The MCP server architecture provides:

OAuth2/OIDC Authentication

Full integration with enterprise identity providers means Claude authenticates as you. Your existing security policies, user entitlements, and access controls all apply. Every query Claude makes carries your authenticated identity.

Complete Audit Trails

Every query is logged with user context. You know exactly who queried what, when, through which interface—whether it was Claude, a direct API call, or the web UI.

Claude-Optimized Responses

We tuned the MCP server specifically for AI consumption:

• JSON streaming instead of binary formats
• 200-row pagination to protect context windows
• Schema enrichment with those SME descriptions

Real Impact at CloudQuant

We've been using this internally. Here's what changed:

Data Discovery

New team members explore datasets conversationally. Claude reads the descriptions and explains what's available in plain language. No more reading documentation or asking around.

Self-Service Analytics

Analysts ask questions without knowing exact column names or table schemas. Claude understands the descriptions and translates natural language into proper queries.

Ad-Hoc Analysis

Questions like "compare volatility across these three datasets last month" become simple conversations instead of writing complex SQL across multiple systems.

Rapid Prototyping

Researchers test hypotheses in natural language before writing production code. The feedback loop goes from hours to minutes.

One engineer described it as: "Having a senior analyst who never sleeps and has perfect memory of every dataset."

Beyond Data Access: Pattern Discovery

The most interesting part? Claude can query multiple datasets and identify patterns humans might miss.

"Are there any unusual correlations between customer demographics and product preferences in the last quarter?" Claude can explore this across datasets, combine results, and present insights you weren't specifically looking for.

This Works Everywhere

While we built this in finance, the pattern applies universally:

• Manufacturing: "Show me production lines where quality metrics dropped after maintenance windows."
• Healthcare: "Find patients with diagnosis code X who had lab result Y within 30 days."
• Retail: "What products are frequently bought together across all our channels?"
• Energy: "Correlate outage events with weather data and grid load."
• Any domain with complex data and people who need to explore it.

The all-too-common scenario at the top of this post plays out every day in organizations sitting on more data than they can actually use. Data Liberator and Claude change that. Your data stays where it lives, your security stays intact, and your team stops waiting. Part 3 of the series is coming soon — but if you're tired of waiting for your data, you don't have to wait for a demo, talk to us.