What it does
The Model Context Protocol is an open standard for exposing tools and data to a model. An MCP server advertises a set of callable operations, the agent discovers them, and from then on querying your database or opening a ticket is a tool call rather than something you paste into chat.
This is the difference between an agent that can only read your repository and one that can check the schema it is writing a migration for, read the issue it is implementing, or look up the runbook it is meant to follow. Cursor runs a marketplace of installable servers, and teams on Teams and Enterprise plans can publish private ones internally.
Access control matters here more than anywhere else in Cursor. Enterprise plans include MCP access controls precisely because an MCP server is, by design, a pipe from an autonomous agent to a production system.
Using it
- Find a server. Browse the Cursor marketplace, or point at any MCP server you or a vendor runs.
- Install and authenticate. Most servers need credentials. Supply them once; the agent then calls the server's tools without seeing the secret in the transcript.
- Scope it. Give the server the narrowest credentials that still let it do its job, and prefer read-only tokens for anything touching production.
Checked against MCP documentation on 25 September 2026.