When a CPA team hears "AI with access to your tracker and chats", the first reaction from leadership is "wait — where does our data go?". It's the right question: partner databases, rates and financial data are literally the business of a CPA network. This article explains what MCP is and why, architecturally, it's a different story from "paste the spreadsheet into a chatbot".
Where MCP came from
The Model Context Protocol (MCP) is an open standard introduced by Anthropic in November 2024. Its job is to standardize how AI systems connect to external data sources and tools: databases, CRMs, trackers, messengers.
The standard quickly became industry-wide: in 2025 it was adopted by OpenAI, Google and Microsoft, and in late 2025 Anthropic donated MCP to independent governance — the Agentic AI Foundation, a fund under the Linux Foundation co-founded by Anthropic, Block and OpenAI. In other words, it's not one vendor's proprietary technology but an open standard used across the industry.
How the MCP approach differs from "dropping a file into ChatGPT"
When a manager copies a tracker export into a public chatbot, here's what happens: the data leaves your infrastructure, lands on someone else's servers, and you lose control over where it's stored and how it's used.
With MCP the logic is reversed:
Data stays where it is
The tracker, BigQuery, chat history — everything lives where it always lived: in your infrastructure. Nothing is bulk-copied anywhere in advance.
AI queries through a controlled connector
When a manager asks a question, the assistant makes a specific request to a specific system through an MCP connector — and receives only the data needed for the answer. Each connector has explicitly defined permissions: what it can read, what it can change.
Encryption and audit
Data is encrypted in transit and at rest. System requests can be logged: you can see who asked what, and when.
An honest caveat about security
MCP is a protocol, not a magic shield. Real security depends on how the system is deployed: where the assistant is hosted, what permissions connectors are given, who has access. That's exactly why, for teams with strict requirements (iGaming, gambling and other regulated verticals), we offer a self-hosted option: the entire system runs on your servers, with your API keys, and data physically never leaves your perimeter.
What this means for a CPA network
Practically: your team gets an assistant that answers questions across the tracker, chat history and registries — while the partner database doesn't "move" to a third-party service, access is granular, and for regulated verticals there's a deployment that stays entirely inside your perimeter. That's the difference between "using AI" and "handing your data to AI".
Want to see how this works in your business?
Book a free Automation Mapping session — we'll analyze your processes and show where AI can save the most time.
Book for free →