What Is MCP? The Model Context Protocol Explained for Business

MCP, the Model Context Protocol, is an open standard for connecting AI assistants to business systems such as a CRM, a shared drive or a database. Instead of building a custom integration for every pairing of AI and tool, a tool offers an MCP server once and any compatible assistant can use it. It is how AI reaches your data.

What is MCP?

MCP is a shared language for AI tools and business software. A system, such as a CRM, offers an MCP server describing what it can do: search contacts, read deals, create a task. An AI assistant that understands MCP can then use those functions on a person's behalf, within the permissions it has been given.

Anthropic introduced the Model Context Protocol on 25 November 2024 as "an open standard" for secure, two-way connections between data sources and AI-powered tools. A common analogy is a universal adapter: one plug that fits many devices, rather than a different cable for each.

In December 2025 Anthropic donated MCP to the Agentic AI Foundation, a fund under the Linux Foundation co-founded with Block and OpenAI. That moved stewardship of the standard to a neutral body.

What problem does MCP solve?

MCP solves the integration problem. Before it, connecting an AI assistant to each business tool meant a separate, custom integration for every pairing. Anthropic's announcement put it plainly: "Every new data source requires its own custom implementation, making truly connected systems difficult to scale." MCP replaces that with one standard.

For a business, the practical effect is that connecting AI to existing tools gets easier and more portable. A connection built on MCP can work with several assistants, so changing or adding an AI tool does not mean rebuilding every integration.

How does MCP work, in plain terms?

There are two sides. A tool provides an MCP server that lists its functions and the data it can share. An AI assistant acts as the client, connecting to that server with a user's authorisation. When a request needs the tool, the assistant calls the right function, gets the result and uses it in its answer or action.

A simple example: you ask your assistant for this week's new deals. Through the CRM's MCP server, it calls a "list deals" function, receives the records, and summarises them. If the server also offers "create task", the assistant could add a follow-up task, provided it has permission.

That last condition matters. MCP describes what can be done; it does not decide what should be. The guide to MCP security covers how to set those limits.

Who supports MCP?

MCP is now widely supported. According to Anthropic's December 2025 announcement, MCP had been adopted by ChatGPT, Cursor, Gemini, Microsoft Copilot and Visual Studio Code. There were more than 10,000 active public MCP servers, and over 75 Claude connectors were powered by MCP.

Business software vendors increasingly offer their own MCP servers. HubSpot, for example, documents a remote MCP server that powers its connectors for several AI assistants. Check each vendor's documentation for what its server can do, because capabilities vary.

What does MCP mean for a business?

It means AI can work with your actual data, not just what someone pastes into a chat. Assistants can read the CRM, search the drive and update project tools directly. That makes AI far more useful, and it makes decisions about access far more important, because connected AI can act as well as read.

Three practical implications:

1. Choose connections deliberately. Connect the systems your frequent AI jobs need, not everything.
2. Start read only. Add write access one action at a time, with a person approving changes.
3. Keep facts separate. MCP gives access to data; your approved facts and methods still need a shared home, such as a company brain.

What are the risks of MCP?

The main risks are connections with more access than they need, malicious or tampered tools, and AI following instructions hidden in data it reads. OWASP's MCP Top 10, currently in beta, lists risks including privilege escalation via scope creep, tool poisoning, prompt injection and shadow MCP servers nobody approved.

Most of these are managed with ordinary discipline: approved servers only, least-privilege access, approval before any external action, and a record of what is connected. According to OWASP, loosely defined permissions "often expand over time, granting agents excessive capabilities", so connections need periodic review.

Skein connects AI to business tools read only first, granting write access one tool and one action at a time; the shared knowledge layer page explains how connections fit the wider system. That suits a business ready to give AI access to its CRM, drive or project tools. A team using AI only for writing may not need MCP connections yet.

Frequently asked questions

What does MCP stand for in AI?

MCP stands for Model Context Protocol. It is an open standard, introduced by Anthropic in November 2024, for connecting AI assistants to external data sources and tools, so an AI can read from and act in systems such as a CRM or shared drive through one common interface.

Is MCP only for Claude?

No. Anthropic created it, but MCP is an open standard now stewarded by the Linux Foundation's Agentic AI Foundation. According to Anthropic, it has been adopted by ChatGPT, Gemini, Microsoft Copilot, Cursor and Visual Studio Code, among others, so one connection can serve several assistants.

What is an MCP server?

An MCP server is the part a tool provides to make its data and functions available to AI assistants. It lists what the tool can do, such as searching records or creating tasks, so any compatible AI assistant can use those functions with a user's permission.

Is MCP secure?

MCP itself is a connection standard; security depends on how connections are set up. OWASP lists risks such as tool poisoning, prompt injection and permission creep. Use approved servers only, give least-privilege access, start read only and require a person's approval for external actions.

Does a small business need MCP?

Only when AI needs to work with live data from business tools, such as CRM records or files in a drive. A team using AI only for writing and analysis on pasted material can manage without it. When you connect, start with one tool, read only.

Related reading