About this project

mcp-vst-sidechain is a Model Context Protocol (MCP) bridge designed to give AI agents full control over VST3 and AU plugins. It acts as an intermediary between an agent and a plugin host, allowing agents to enumerate, read, and set automatable parameters using real-world units rather than raw normalized values. Key capabilities include: - Semantic Parameter Mapping: A layer that recovers the meaning, units, and curves of parameters, persisting this knowledge across sessions so agents can reason about roles (e.g., "filter cutoff") rather than opaque IDs. - Closed-Loop Tuning: The system can render plugin output offline and objectively measure characteristics such as spectral centroid (brightness), RMS, and modulation. This allows agents to autonomously tune parameters to reach a specific sonic goal (e.g., "make it brighter"). - Multi-Controller Support: Multiple agents can drive a single plugin instance concurrently using an edit-lease system to prevent conflicting changes. - Efficient Data Transfer: Uses GCF (a compact wire format) to reduce the token cost of serializing large parameter sets compared to standard JSON. - Broad Compatibility: Supports VST3 (cross-platform) and AU (macOS) via the JUCE framework. The architecture consists of a Go-based MCP server for agent communication and a C++/JUCE host for plugin management, communicating via a localhost TCP socket.