Practical MCP with FastMCP & LangChain

Function Calling vs MCP Tools

Function calling and MCP tools can look very similar because, in both cases, a model "calls a tool". But they operate at different levels and serve different purposes.

When you use function calling, you tell the model which tools are available by including their definitions directly in your request. The model can then respond with structured arguments saying, in effect, Please run this function with these inputs. Your application receives that request, runs the function in your own code, and sends the result back to the model. Everything happens inside your application. You control the tool definitions, the execution logic, the credentials, and the error handling.

MCP tools are part of a broader protocol between a client and a server. Instead of defining tools inside each model request, the tools live on an MCP server. A client connects to that server, asks what tools are available, and then calls them through a standardized mechanism. The model may still decide to use a tool, but the tool itself is not embedded in your API request. It exists independently on the server.

With function calling, the model talks directly to your application’s code. With MCP tools, the client talks to a separate tool server using a shared protocol. That difference becomes important when you think about portability and reuse. If you switch model providers, function calling setups often require changes in your application. MCP tools, on the other hand, remain exposed in the same way because the protocol does not depend on a specific model vendor.

There is also a difference in discovery