Can an MCP App trigger the host model privately without adding a host chat turn?

[ideafoundryai]

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

We’re building an interactive MCP App where a user types a natural-language instruction inside the App (e.g. “make this shorter”), the host model should act on private App context, and the result should return to the App — with nothing added to the host conversation UI/transcript.

We’ve confirmed:

updateModelContext successfully provides private App state/context to the host model, but does not trigger inference.
sendMessage can drive the host forward, but creates a visible conversation turn.
createSamplingMessage would provide direct model execution, but sampling is not currently advertised by either Claude or ChatGPT in our MCP App hosts.
App-initiated callServerTool() is proxied to the MCP server and returns the result to the App, but does not appear to start/resume/re-enter the host model execution loop.

Manual user interaction inside the App is completely fine — e.g. type an instruction and click Refine.

What we’re looking for is:

App-local user action → host model executes using private App context → result returns to App

while:

no sendMessage / no visible host turn / no model response in host chat

Is there a supported MCP Apps pattern, user-gesture lifecycle, host capability, or host-specific mechanism we’re missing that enables this today?

We specifically want to use the host model, not run separate model inference on the MCP server.