OpenAI function calling vs MCP
Function calling is great inside one model API. MCP aims to keep tools portable across runtimes.
Function calling (and similar tool APIs) let a model request structured actions. That works well when you stay inside one provider SDK.
Portability
MCP separates the tool server from the model vendor. You can change models without rewriting every tool adapter.
Operations
A shared MCP control plane gives auth, audit, and metering that sit outside any single model bill.
Use the model for reasoning. Use MCP for shared tools.