AI and agents
Model Context Protocol (MCP)
Model Context Protocol (MCP) is an open standard, introduced by Anthropic in November 2024, for connecting AI applications to tools and data through a common interface called an MCP server.
How it works in practice
Before MCP, each AI app needed its own connector for each system. An MCP server describes what a system offers, mainly tools the model can call, resources it can read and prompts it can use, in a format any MCP client understands. Claude, ChatGPT, Cursor and many other clients can connect to the same server, so one integration serves all of them.
For a business, an MCP server is a controlled door into your systems. You decide which actions it exposes, what each one may change and who can connect. A server that wraps your CRM with read-only tools is low risk; one that can send email or move money needs authentication, scoped permissions and an approval step. MCP does not handle those decisions for you, so they belong in the design.
A worked example
An accountancy firm wants staff to ask Claude about client deadlines. We build an MCP server over their practice system with three read-only tools: find a client, list upcoming filings and show outstanding documents. Staff connect it in their AI client, and nothing in the practice system can be changed through it.
Examples in this glossary describe typical builds for companies of 1 to 1,000 people. They are not named client results.















