Model Context Protocol, explained
MCP gives different AI applications and outside services a common way to describe capabilities and exchange requests. It reduces the need for a separate connection design for every pairing.
Model Context Protocol, or MCP, is an open standard that connects AI applications with external tools, data, and services through a shared client-server format.
MCP separates the application from the service
The host is the AI application a person uses. A client inside that host communicates with an MCP server. The server exposes approved tools, information, or reusable prompts.
This separation lets a service present the same MCP connection to more than one compatible host. Access controls and permissions still belong around that connection.
The protocol does not run the workflow
MCP standardizes communication. It does not decide which business goal to pursue, when an agent should call a tool, or whether a proposed action should be approved.
The host and its surrounding controls make those decisions. MCP carries the structured request and result between the parts.
Model Context Protocol terms
- MCP host
- The AI application where a person makes a request and the model works.
- MCP client
- The part inside the host that manages communication with one MCP server.
- MCP server
- A service that exposes approved tools, information, or reusable prompts through the protocol.
What MCP standardizes
Tool discovery
A client can learn which approved capabilities a server provides.
Structured messages
Clients and servers exchange requests and results in a shared format.
Clear separation
The protocol connects the parts, while the application decides how and when to use them.
Questions about Model Context Protocol
No. It is a connection standard used by AI applications. The model, host, client, and server remain separate parts.
No. Tool calling is the act of proposing and handling a tool request. MCP standardizes how applications connect to services that can expose those tools.
No. The application still needs trusted servers, limited permissions, input checks, and approval rules for important actions.
Related glossary terms
Connect tools with clear boundaries
Caffeine apps run on the Internet Computer, a public network built for this.