What is an MCP server for corporate data?

MCP in one sentence

Model Context Protocol, abbreviated MCP, is an open protocol that allows an AI application to find and use external context and tools. A host such as ChatGPT or Claude includes an MCP client. That client connects to a server that offers clearly described capabilities.

This separation is especially important for company data. The assistant does not automatically get access to an entire database and does not need to know a source API. The DigiData MCP Server provides a small, fixed collection of read-only tools for sources, tables, schemas, bounded queries, aggregations, and sync state.

What a DigiData administrator determines

An administrator creates a connection in DigiData and selects the allowed source, table and columns. The tenant comes from the validated user and connection; the model cannot choose a tenant ID. DigiData checks the current connection and allowlist again with every call.

That is different from uploading a file. A file is a snapshot that can persist outside DigiData. MCP can request the last synchronized, explicitly allowed data, but only within server-side limits. An administrator can change the selection or revoke the connection.

What tools are available?

The discovery begins with sources and diagrams. Then a client can read selected columns or perform an approved count, sum, average, or grouping. There is no add, change, delete, send or approve tool. There is also no free SQL input.

A safe starting pattern is therefore: first ask what resources are available, check the synchronization status, describe the desired table and only then request a small result. The MCP Troubleshooting page describes the same fault diagnosis sequence.

How authentication works

Remote MCP used for DigiData OAuth with PKCE. The remote client does not receive an OData API key. During the consent, the administrator sees which connection and dataset are being shared. Tokens are tied to the intended DigiData resource; revocation and the current connection status are maintained server-side.

Read the technical security choices at MCP security. It also explains why a cloud provider does not automatically fall under the IP allowlist of a browser or Power BI user.

MCP or DigiData AI?

MCP is intended for organizations that consciously want to use an external assistant. DigiData AI is the recommended route for recurring analytics with managed definitions, approved metrics, dashboards, Teams, email, and automations. The comparison between DigiData AI, ChatGPT and Claude makes that choice explicit.

Privacy boundary

DigiData only returns the permitted data and does not use customer questions or company data to train models. Once an administrator has data sent to an external assistant, that provider processes its own copy under the chosen plan, contract and management settings. So check that provider layer next to the DigiData permission.

Summary

An MCP server makes AI links repeatable and writable, but the protocol does not replace access policies. The value of DigiData MCP lies in the combination of pre-synchronized data, OAuth, current tenant control and a limited read-only toolset. Start on the MCP Hub or follow the provider guide for ChatGPT or Claude.