AI What is the Model Context Protocol (MCP)?

The Model Context Protocol (MCP) is an open standard, introduced by Anthropic in late 2024, for connecting AI applications to the tools and data they need. Before MCP, every integration between an AI assistant and a business system was custom work. With MCP, a system exposes its capabilities once, through an MCP server, and any compatible AI client — assistants, coding tools, custom agents — can use them.

A common comparison is a universal port: instead of a different cable for every device, one standard connector. The analogy is imperfect, but it captures why MCP spread so quickly across AI products and why it matters for any business that wants AI to work with its own systems.

The parts of MCP

• Host or client: the AI application — a desktop assistant, an IDE, or your own agent — that connects to servers.
• Server: a program that exposes capabilities from a system, such as a CRM, database, file store or internal API.
• Tools: actions the AI can call, each with a name, a description and typed inputs.
• Resources: data the AI can read, such as files or records.
• Prompts: reusable templates a server can offer for common tasks.

When a client connects, it asks the server what tools and resources it offers. The model then decides when to call a tool and with what inputs; the client runs the call through the server and returns the result to the model.

Local and remote servers

MCP servers can run locally, on the user's machine, which suits developer tools that work with local files. Or they can run remotely as a web service, which suits business systems used by many people. Remote servers use OAuth for authorization, so the AI acts on behalf of a specific signed-in user and is limited to what that user — and the scopes they grant — allow.

What MCP means for your business

If you want AI assistants or agents to work with your systems, an MCP server is now the standard way to do it. If you sell software, customers increasingly ask whether your product has an MCP server so they can use it from their own AI tools. In both cases the server is the controlled doorway between AI and your data.

The parts that need care

The protocol itself is the easy part. The decisions that determine whether connecting AI to your systems is safe are about permissions and operations:

• Which tools to expose, shaped around tasks rather than every API endpoint.
• Scopes per tool, with read separated from write and high-impact actions gated behind approval.
• Separating servers by trust boundary, so one server cannot both read sensitive data and send it outside.
• Rate limits and clear errors when a downstream system is slow or refuses.
• An audit log of every call: who, what was sent, and what came back.

MCP and other protocols

MCP connects AI applications to tools and data. A separate open protocol, Agent2Agent (A2A), addresses how agents from different teams or vendors communicate with each other. The two are complementary: an agent might use MCP to reach a CRM and A2A to hand a task to another organisation's agent.

Getting started

If you already have an internal API, an MCP server in front of it is usually a modest project — a few weeks to design the tools, add authorization and logging, and test with real assistants. Spend the time on the permission model: it decides whether AI access to your systems is a feature or a liability.