Is MCP always a make-request-get-answer protocol, or does it allow for multiple back-and-forth messages with the tool being called?

[stebansaa]

Pre-submission Checklist

• I have verified that this discussion would not be more appropriate as an issue in a specific repository
• I have searched existing discussions to avoid duplicates

Discussion Topic

I'm trying to better understand what can be done with MCP, from my current understanding the LLM triggers a tool call that returns information indicating the competition of a process, then this is info returned is part of the context the LLM needs to understand to pass information back to the user.

But what happens when a more "active" communication needs to happen, that is the info returned by the tool calling, needs to be processed and then again returned to the tool. An internal conversation between the two parties. Is this already possible, or outside the scope of project?

Thank you

[olaservo]

When you say "needs to be processed and then again returned to the tool" who or what is doing the processing in this scenario (and/or what are the two parties having the internal conversation)?

Wanted to clarify since "processing" here could mean:

• the LLM processing the tool's response and then making another tool call
• the end user/human in the loop processing the tool's response and providing additional input
• another system component that sits between the LLM and the tool
• a direct back-and-forth between the tool and some other entity without the LLM as intermediary

So I think it would help to know which of those scenarios you're talking about.