What is MCP? Explained with an office of service counters
MCP (Model Context Protocol) is one shared way for AI assistants to connect to other software and use it with your permission. Explained with an office where every department takes the same request slip.
- Author
Umair Mateen- Published
- Reading time
- 6 min
- Category
- AI
Webull Canada is finishing a rollout that lets customers link their brokerage accounts to ChatGPT, Claude, Perplexity and Grok, and the announcement says the link runs on the Model Context Protocol. Here is what that protocol is, in plain words.
In one sentence: the Model Context Protocol (MCP) is an open standard that lets an AI assistant connect to other software, such as a calendar, a file store or a brokerage account, in one shared way, so it can look things up and take actions there with your permission.
The everyday picture
Imagine a large office with many departments: accounts, bookings, the archive, the warehouse. You have a personal assistant who runs errands to all of them for you.
In the old office, every department had its own forms and rules, and each new one meant new training. MCP is like the whole building agreeing on one way of working. Every department puts up a menu at its counter, listing what it can do and what details it needs. Every request goes on the same standard slip.
Now the assistant can use any counter: read the menu, fill in a slip, done. A new department just sets up a standard counter, and every assistant can use it.
The AI app reads a service's menu of tools, fills in MCP's standard request, and the service's MCP server does the work. The same format works for every connected service.
How it works, step by step
- You ask the AI app (your assistant) for something in plain words, such as "What meetings do I have on Friday?" In MCP terms, the AI app is called the host.
- The app finds the right counter. Each connected service runs an MCP server (a department counter). The app keeps a separate line open to each one, called an MCP client. It asks each server for its list of tools (the menu), with a description of each and the details it needs.
- The app fills in the standard slip. The AI model picks a tool and writes a request in MCP's fixed format, which is the same whichever service it is talking to. The specification says the app must get your clear consent before it uses a tool.
- The counter does the work. The MCP server turns the slip into real work inside its own system, such as reading your calendar or preparing an order. Servers can also share resources (documents and data to read) and prompts (ready-made instructions).
- The answer comes back on the same kind of slip, and the AI app tells you the result in plain words.
Where the picture stops
- The assistant is not a person. An AI model decides which tool to use and what to write on the slip, and it can misunderstand. Webull's own notice warns that AI tools may misread instructions, which is why trades only go through after the investor confirms them on Webull.
- A counter can lie about its menu. In a real office you trust the departments. With MCP, anyone can build a server. The specification warns that tool descriptions should be treated as untrusted unless they come from a trusted server, so only connect servers from sources you trust.
- The standard slip does not enforce the rules. MCP sets out that people must approve actions and control what data is shared, but the specification admits the protocol cannot enforce this by itself. Each app has to build those approval steps.
Why it matters for your business
- Build once, reach many AI apps. If your product offers an MCP server, it can work with any AI app that supports MCP. The MCP project has named ChatGPT, Claude, Cursor, Gemini, Microsoft Copilot and Visual Studio Code among the apps that support it.
- Customers may arrive through their assistant. As the Webull news shows, people are starting to use services by asking an AI app instead of opening a website.
- New risks to manage. Every connected server is a door into your systems. Give each one the least access it needs, and keep a person approving actions that move money or data.
An example
Imagine a 25-person property management office whose staff keep checking a booking calendar and a shared drive of leases.
The office connects its AI assistant to two MCP servers: one for the calendar and one for the document store. A staff member asks, "When is the next viewing for Flat 4, and when does its lease end?" The assistant checks both and answers in one reply, linking the lease. Asked to move the viewing, it shows the change and waits for approval before touching the calendar.
Frequently asked questions
What does MCP stand for?
MCP stands for Model Context Protocol. "Model" is the AI model, "context" is the information and tools it is given, and "protocol" means an agreed set of rules for how two pieces of software talk.
Is MCP only for Claude?
No. Anthropic introduced MCP in November 2024, and in December 2025 it handed the project to the Agentic AI Foundation, part of the Linux Foundation. It is an open standard that many AI apps from different companies support.
Is MCP the same as an API?
Not quite. An API is the doorway one piece of software offers to others, and each API has its own format. MCP is one common format designed for AI apps, and an MCP server often uses a service's normal API behind the counter.
Is it safe to connect my accounts through MCP?
It depends on the app and the server, not on MCP alone. Use servers from the official provider or another trusted source, and check that the app asks before acting.
Related explainers
- What is an API? Explained with a waiter in a restaurant
- What is an LLM? Explained with a new hire who read the whole library
- What is RAG? Explained with an open-book exam
The DevDrift connects AI assistants to business systems through MCP, with approvals and audit trails: see how.