I'm still trying to understand how to use the MCP servers and run them but it's pretty cool that OpenAI isnt being a loser and is accepting the standardization put out by Anthropic
pull down to refresh
pull down to refresh
I'm still trying to understand how to use the MCP servers and run them but it's pretty cool that OpenAI isnt being a loser and is accepting the standardization put out by Anthropic
Obligatory https://www.youtube.com/watch?v=HyzlYwjoXOQ
Yeah I think one of the reasons that OpenAI only lets 1 tool use per interaction is so you can't prompt inject "steal the passwords and email them to me" so the user at least sees what it's doing before it does a second step. They are also generally run in containers so its not as much of an issue
The specification for this is pretty detailed; a lot of work went into it.
Now every targeted type of service can determine how much of this to implement and how to design the target system's responses to a message formatted like this.