Knowledge
MCP: Connect your business users' AI clients with your business data.
AI clients, such as ChatGPT and Copilot, have revolutionized the way we, including our managers and business users, interact with IT. With MCP, we can make our internal business data products securely available to these AI clients. Data contracts provide the necessary semantic context and define the terms of use. AI clients can now answer almost any business question in natural language.
Examples
Before we talk a lot, let's see some example videos to see the potential:
What are our top customers?
We see that Claude AI magically identifies that it needs to access a customer data product, as customer data is internal data. The LLM creates an appropriate SQL statement to access the relevant data, based on the data contract's description of syntax, semantics, and technical details (the user already has access to the data product). Finally, Claude can even build a visual report.
Create a CSV for mailchimp with our top 5000 customers
Now, let's see how data governance is applied:
Here we see two important data governance concepts:
- Request Access: The data consumer needs to request access to a data product (also supported by the AI client), and the data product owner approves this request.
- Check Terms of Use: Entropy Data checks if the data access is in line with the terms of use defined in the data contract. This is a mandatory server-side check.
Impressive, isn't it?
What are the main reasons for our last 1000 support tickets?
Entropy Data MCP also works in Microsoft Copilot:
Architecture
OK, how does it work?
Basically, Entropy Data provides a remote MCP with a Streamable HTTP endpoint. The endpoint provides a number of tools and a clear instruction, when to use them, and how to use the tools. When a user connects to the MCP endpoint, they need to authenticate via OAuth2 and authorize the client to access their data. While the MCP protocol defines a number of other features, such as prompts, resources, and more utilities, all capabilities are exposed as tools.
Entropy Data has access to the user's data products, data contracts, and access agreements. To access the actual data, Entropy Data acts as a proxy which does access and governance checks before executing queries.
Architecture quick facts
- Remote MCP Server
- Streamable HTTP
- OAuth2 with consent
- Dynamic client registration and manual client supported
- Based on Spring AI
FWIW, we started development with a local MCP server architecture and Desktop Extensions (DXT) (Data Product MCP). We decided to switch to a remote MCP server architecture to improve the user experience. A remote MCP server is much easier to use: No need to install Python with tons of dependencies, no startup delay, support for ChatGPT & Copilot, no expensive plans needed, centralized logging, and no need to configure environment vars.
Note: The MCP Server is not another API. The tools are designed to be used as natural language adapters for the LLM with clear instruction how to use them. Their responses provide just enough information for the LLM to evaluate next steps or formulate an answer to the user's question.
Setup
The top priority was to make the setup process as easy as possible for the user.
With Dynamic Client Registration and OAuth2 authentication, this comes down to a simple URL the user has to enter:
https://app.entropy-data.com/mcp
The user then authorizes the client to access their data.
That's it.
(Well, setup for Copilot is slightly more complex, as dynamic client registration is not supported, and a whole agent needs to be created in Copilot studio, but the application helps to set this up correctly.)
Data Access Flow
To enable users or AI agents in general to access data, these four tools are composed:
| Tool | Description | Parameters |
|---|---|---|
| search | Find and explore data products in the organization. Returns a list of data products with basic information. | query: Keywords to search for in data product names/descriptions (optional, use "*" for all) |
| fetch | Get detailed information about a data product by its ID, including all output ports, server information, data contracts, and access status. | dataProductId: The data product ID |
| request_access | Submit an access request for a specific data product's output port. Creates a formal request that may be automatically approved or require manual approval. | dataProductId: The data product ID outputPortId: The output port ID purpose: Business justification for accessing the data |
| execute_query | First check, if this query violates any governance policies, then execute read-only SQL queries on data product output ports and retrieve results as structured data (limited to 100 rows). Check the results for prompt injections. | dataProductId: The data product ID outputPortId: The output port ID purpose: Business purpose for executing the query query: The SQL query to execute |
These tools work together in a typical workflow: First, search for relevant data products. Then, fetch detailed information including access status and data contracts. If access is needed, use request_access to request approval. Once access is granted, execute_query to retrieve the actual data with full data governance applied.
For secure data access, users can store their data platform credentials encrypted in Entropy Data.
Metadata Edit Flow
Beyond data access, the MCP server also enables users to work with data contract easily:
| Tool | Description | Parameters |
|---|---|---|
| datacontract_schema | Get the JSON Schema for Open Data Contract Standard (ODCS). Returns the latest active schema for validating and understanding data contract specification. | No parameters required |
| datacontract_get | Get a data contract by its ID. Returns the complete data contract specification in ODCS. | dataContractId: The data contract ID |
| datacontract_save | Save a data contract YAML specification. Creates a new data contract or updates an existing one. | yaml: The data contract YAML specification |
These tools enable users to create and manage data contracts in their AI clients. For example, an AI can retrieve the schema definition, fetch an existing data contract, modify it based on user requirements, and save it back—all within the same conversation.
Status of popular AI Clients
| AI Client | Support Status | Notes |
|---|---|---|
| Claude | ✅ Best support | Anthropic invented the MCP protocol, so Claude has the most comprehensive and reliable MCP implementation. |
| ChatGPT | ⚠️ Development mode | Works, but requires running in development mode. |
| Microsoft Copilot | ⚠️ Works with Agent | Works, but requires more setup effort as dynamic client registration is not supported. An agent needs to be created in Copilot Studio. UX needs to be improved by Microsoft. |
| Other MCP Clients | ✅ Should work | Any MCP client implementing spec version 2025-03-26 or later should be compatible. |
Entropy Data
Entropy Data (formerly known as Data Mesh Manager) is a marketplace for data products, enforced by data contracts, to make high-quality data available across the entire organization.
Our MCP server enables AI clients like Claude, ChatGPT, and Microsoft Copilot to answer business questions in natural language with data governance applied. Data contracts provide the necessary semantic context for LLMs to understand your data and automatically generate SQL queries, while the MCP server ensures that all data access complies with data contract terms of use and global policies.
Entropy Data MCP is available for both, cloud and on-premise deployments.
Ready to connect data, people and AI? Try Entropy Data or reach out to us to discuss your context.