MCP, the Model Context Protocol, is an open standard for connecting AI applications to other systems. It's how an AI tool such as Claude reaches the files, databases and business software outside its own chat window, using one common method for all of them. The project's own documentation compares it to a USB-C port for AI applications.
This page explains it in plain terms, from the official MCP documentation and Anthropic's help pages, checked on 4 November 2026.
What MCP does
An AI model on its own only knows what's in the conversation. To read a spreadsheet, look something up in a CRM or create a calendar entry, it needs a connection to that system. MCP gives AI applications and outside systems a shared standard for that connection, so one MCP server can work with any AI application that supports the protocol.
Hosts, clients and servers
MCP has three parts:
- The host is the AI application you use, such as Claude Desktop or Claude Code. It manages one or more connections.
- The client sits inside the host and keeps the connection to one server.
- The server is a program that gives the AI access to one outside system, such as a file store, a database or a piece of business software.
A server can offer three kinds of thing: tools, which are actions the AI can run; resources, which are data it can read; and prompts, which are reusable templates. Servers run either on your own computer, talking to the host directly, or on the internet over a transport the protocol calls Streamable HTTP.
MCP servers in Claude
- Claude, Cowork and Claude Desktop. Anthropic calls MCP connections "connectors". Custom connectors using remote MCP servers are available from the Free plan up to Enterprise, with Free limited to one custom connector. A remote server has to be reachable over the public internet.
- Claude Code. Servers are added from the terminal with
claude mcp add. For a remote server:claude mcp add --transport http <name> <url>. A server can be added for one project or for every project on the machine.
The details are in Anthropic's Claude Code MCP docs and its custom connectors article. Cowork itself is covered on the Claude Cowork guide.
Who governs MCP
Anthropic introduced MCP in late 2024. In December 2025 it donated the protocol to the Agentic AI Foundation, a directed fund under the Linux Foundation co-founded by Anthropic, Block and OpenAI. Anthropic said at the time that the protocol's governance model would stay the same.
What to check before connecting a server
- Who made it. Anthropic's docs say to verify you trust each server before connecting it, and custom connectors aren't verified by Anthropic.
- What it can reach. A server that runs on your own computer runs with the same access as the AI application using it.
- Outside content. Servers that fetch web pages or other outside content can expose the AI to prompt injection, where instructions hidden in that content try to steer it.
- What it asks before acting. The MCP security guidance says applications should ask for consent before a local server runs commands.
Where MCP fits in a business
For a business, MCP is one of the ways an AI agent gets access to the software the business already runs. The agents on AI agent setup are connected to what the business already uses: the accounting export, the inbox, the spreadsheets and the website. Where the agent builds custom apps to replace subscription software, that's agent services for businesses. The Claude Code course sets up an agent on your own machine in half a day.
Working out what to connect.
Email what your business runs on and what you'd want an agent to reach, and you'll get a straight answer on how it fits.
Get in touch →Questions about MCP
What is MCP?
MCP, the Model Context Protocol, is an open-source standard for connecting AI applications to external systems such as files, databases, business software and other tools. Its own documentation compares it to a USB-C port for AI applications.
What is an MCP server?
A program that gives an AI application access to one outside system. It can offer tools the AI can run, resources it can read and prompt templates. The AI application connects to it through an MCP client.
Who owns the Model Context Protocol?
Anthropic introduced MCP and, in December 2025, donated it to the Agentic AI Foundation, a directed fund under the Linux Foundation co-founded by Anthropic, Block and OpenAI.
Can I use MCP with Claude?
Yes. Custom connectors using remote MCP servers work in Claude, Cowork and Claude Desktop, with Free plans limited to one custom connector. Claude Code adds servers with the claude mcp add command.
Is MCP safe for a business to use?
It depends on the server. Anthropic's docs say to verify you trust each server before connecting it, note that custom connectors are not verified by Anthropic, and warn that servers which fetch outside content can expose you to prompt injection.