Modifying LLM response output with tools

[kappagamma25]

Question

Iis it possible to edit the response from the LLM before it is presented to user? For example, I obfuscate some of the details of an API response to ensure confidential data is not sent to the LLM (e.g. masking some username or otherwise) with a placeholder, but then would like to unmask this placeholder where it occurs in the LLM response as it is presented to the user.

Is there some decorator I can use to apply tools to the LLM output response? Or another solution?

Additional Context

No response

[YsK-dev]

the unmasking you're describing would typically happen in your application logic after you receive the response from the LLM and before you display it to the user. this is outside the scope of the Model Context Protocol itself.

[felixweinberger]

the unmasking you're describing would typically happen in your application logic after you receive the response from the LLM and before you display it to the user. this is outside the scope of the Model Context Protocol itself.

Yup if you have some need for predictable processing of responses this should be done within your application.