I've always considered MCP great for 3rd party integrations, in more of a plugin style model. But tool calling still has uses for developing more complex AI driven algorithms, like giving it an add function or a DB read function and the model can solve more things all in one prompt, and you are okay with tool calls being tightly coupled
Would be great to see an article on PocketFlow and how you'd approach implementing tool calling and LLM driven state transitions to do more complex and non-linear tasks.
Great article, been loving all your posts.
I've always considered MCP great for 3rd party integrations, in more of a plugin style model. But tool calling still has uses for developing more complex AI driven algorithms, like giving it an add function or a DB read function and the model can solve more things all in one prompt, and you are okay with tool calls being tightly coupled
Would be great to see an article on PocketFlow and how you'd approach implementing tool calling and LLM driven state transitions to do more complex and non-linear tasks.
Thank you for your article. It has been of great help to me.