What is model context protocol?
Model Context Protocol, or MCP, is an open standard, introduced by Anthropic in 2024, that defines a common way for AI assistants to connect to outside tools and data sources such as file storage, CRMs and databases.
Before MCP, every connection between an AI assistant and a business system was a custom job. MCP works more like a universal adapter: a tool publishes an MCP server describing what it can do, and any compatible AI client can use those capabilities. Several AI vendors and many software companies have adopted it, so more connectors appear each month.
For a business, the benefit is less custom integration work when you want an assistant to read a shared drive, look up a record or create a task. The risk is the same as with any integration: once connected, the assistant can do whatever the connector allows. A connector with write access to your CRM can change records, not only read them.
Treat MCP connectors like any other third-party integration. Ask who built the server, what permissions it requests, where data travels, and whether actions are logged. Start with read-only access, use a dedicated account with limited rights, and keep an inventory of what is connected. Your AI acceptable use policy should say who may add new connectors.
Example
For example, an architecture firm's project managers could connect an AI assistant to their project folders and task tracker through MCP connectors. A project manager asks for every open RFI on a school project and a draft status note for the client. The connector is read-only, so the manager copies the approved note into email rather than letting the assistant send it.