The audio processing objects in the Max/MSP platform operate on fixed size buffers of audio samples and are executed in a separate audio processing thread from the message handling objects. Implementing audio processing algorithms typically requires direct access to the individual samples. Currently, the two most common ways of accomplishing this are by compiling Max/MSP externals written in C or C++ or using the GEN sample-based visual patching environment that is also compiled. In this paper, we describe an alternative approach that uses an embedded Scheme interpreter that executes in the audio processing thread and enables direct manipulation of individual samples in real-time. The code is interpreted at runtime and can be modified on-the-fly. This is based on the Scheme for Max (S4M) extension to Max/MSP which was originally developed to operate in either the main or scheduler message threads. Three examples of use cases are described: spectral processing, audio effects, and audio feature extraction. The proposed approach enables rapid prototyping of audio and music processing that can be controlled interactively using the Max/MSP patching environment without requiring a compilation stage.