With news of MCP coming eventually for those of us who don’t code in Python, what are some things we could do to get ready for this? Any thoughts?
1 Like
To add context…
To prepare…
Note in the middle of this blog article a familiar endpoint SDK that is being built upon, one with versions in several programming languages.
1 Like
I started adding MCP support to my FastAPI application. Originally using the FastAPIMCP library but ended up ‘doing my own version’ - I don’t need just all API paths exposed, I am much more interested in having all the different tools I have and now always programmatically need to tie to a prompt solved with MCP. It is in essence a super simple protocal with ‘tools’ and ‘resources’ (and templates are pointers to specific resources). You need to register them and you need ‘answer’ the call for a tool or resource. Just much easier to handle tool calls with a server in a standard way. Over time it will be more expansive. I’ll post ‘my’ FastApi solution here soon to share.
2 Likes
To add to the ‘how to prepare’ the concept of MCP servers will probably bring ‘batch processing’ (now only available with Assistants) back to responses in a ‘better than ever’ way. MCP servers should allow OpenAI to let you submit batches of responses with MCP server information for tool calling. That way they can run batches WITH tools calling whenever they want (as long as the MPC server is up and running of course). It will also allow to a much better experience in the Playground - once you are able to connect an MCP server you will be able to truly test stuff with tools easyly.