MCP vs plugins vs function calling: which should you use?
The one-line difference
- Function calling is the model-level mechanism: the LLM emits a structured call and your code runs it.
- MCP (Model Context Protocol) is an open standard for exposing tools/data so any client can use them without bespoke wiring.
- Plugins are packaged capabilities for a specific host or framework.
When to use each
Function calling — you're building one app and want the model to trigger your own functions. Simple, direct, framework-native.
MCP — you want tools that are reusable across many agents/clients, or you want to consume someone else's tools without writing an integration. MCP servers are swappable and standardized.
Plugins — you're extending a specific platform and want a drop-in package that follows that platform's conventions.
They compose
These aren't mutually exclusive. An agent can use function calling to invoke an MCP client that talks to a remote MCP server, while a plugin bundles the whole thing for a host. Pick the lowest-effort layer that gives you the reuse you need.
On ai-supply
MCP servers, plugins, and capabilities designed for function-calling all live in the catalog, each security-scanned. Browse by kind on the categories page and connect in a minute via the quickstart.