MCP stands for Model Context Protocol. It is an open standard, introduced by Anthropic in 2024 and now supported across the industry, that defines how an AI assistant talks to other software.
An MCP server is a small program that sits next to one of your systems and describes it to AI as a list of tools. A CRM connector might offer search_contacts, get_deal and log_note. When someone asks the assistant "what is the status of the Acme deal?", the assistant picks the right tool, calls it, and answers with real data.
Why it matters
Before MCP, every AI product needed its own integration with every tool. With MCP, you build one server and every compatible client can use it: Claude, ChatGPT, Cursor, VS Code agents and agent frameworks.
The three parts
- Host or client: the AI app your team uses.
- Server: the connector to your system. This is what we build.
- Your system: CRM, database, ERP, internal API.
What a server can expose
Mostly tools (actions the model can call), plus resources (data the client can read) and prompts (reusable templates). For business connectors, well-designed tools do most of the work.
Local or remote
A local server runs on each person's computer and is started by the AI client. A remote server runs in the cloud, is shared by the whole team and usually signs users in with OAuth. Remote servers are easier to manage and audit; local ones are quicker to try.
Do you need a custom one?
If a ready-made server covers your system and your fields, use it. Build a custom one when you need your custom objects, tighter permissions, an audit log, or one tool that pulls from several systems. Read how much a custom MCP server costs.