MCP
acronymcloudprogramming
Stands for: Model Context Protocol
An open protocol that lets AI models call external tools and data sources in a standard way.
Model Context Protocol standardizes how a language model connects to tools, files, and services, so an AI application can use many capabilities through one interface rather than a bespoke integration for each.
MCP standardizes how a language model reaches tools, files and services, replacing a situation where every integration was bespoke. A server exposes capabilities in a defined shape, any compliant client can use them, and the same integration works across different models rather than being written once per vendor.
The reason it matters architecturally is that it separates the model from the plumbing. Before, connecting a model to a ticketing system meant writing glue specific to both ends; after, the ticketing system exposes an MCP server once and every client benefits. That is the same argument that made LSP succeed in editors, and the analogy is close enough to be predictive.
The security considerations are the interesting part and are not yet settled practice. A server given broad access is a capability the model can be persuaded to misuse, and prompt injection through returned content is a real path from untrusted data to unintended action. The disciplines that apply are the ordinary ones applied carefully: least privilege on what each server can reach, treating tool output as untrusted input rather than as instruction, and human confirmation for anything destructive or irreversible.
Also known as: mcp