how could we introduce a human-in-the-loop step for double confirm and approval while implement a MCP server / client?

[Colstuwjx]

Pre-submission Checklist

• I have checked that this question would not be more appropriate as an issue in a specific repository
• I have searched existing discussions and documentation for answers

Question Category

• Protocol Specification
• SDK Usage
• Server Implementation
• General Implementation
• Documentation
• Other

Your Question

Hi team, I've checked the document and seen a message from the document here, I'm wondering how we could introduce a human-in-the-loop step for double confirm since some tool invocations might be potential risky to the end users and we want to build an approval mechanism for these invocations.

Any detail implementation guide for this ? For example, langgraph did this by a function like input in Python.

Thanks.

[leo237]

Currently, human validation is done on the client side. Before you send a tool-execution request, you may want to run this by a human in your application.

However, you bring an interesting point about servers potentially being able to trigger this as well -- essentially adding MCP client capability to handle human input requests from the server (i.e. server -> client -> human flow), similar to how sampling/roots works

For example, tools that execute dynamic workflows could benefit from this protocol. Specific steps within the workflow - such as approving critical changes or validating sensitive operations - can require human confirmation, but the overall workflow execution itself might not need validation.