MCP (Model Context Protocol) is an open protocol that standardizes how an artificial intelligence application — an assistant or an agent — connects to external tools, data and systems. Introduced in late 2024, it plays a role for AI comparable to a universal port: instead of building a custom integration for each service, you expose a single interface that any compatible agent can consume.
Technically, MCP relies on a client-server architecture. The MCP server publishes capabilities — tools (actions the agent can perform) and resources (data it can read) — described in a structured way. The MCP client, embedded in the language model, discovers these capabilities and calls them throughout the conversation. Because the exchanges are typed, the model knows which parameters to provide and which responses to expect.
The protocol's value is interoperability: a single MCP server can be plugged into different agents (desktop assistants, development environments, automations) without rewriting the integration layer. Security relies on the usual Web mechanisms — token authentication, often via OAuth — and on the server enforcing the user's permissions.
eyeot exposes an MCP interface that lets an AI agent query and act on the ERP while reusing the same authentication and permission rules as the conventional interface. The protocol thus complements the traditional REST API for AI-driven use cases, without bypassing the existing security controls.