MacromNex/prottrans_mcp

Implements four specialized tools for protein analysis: embedding extraction via ProtTrans language models, mutation scoring with ProtBERT log-likelihood, fitness model training with 5-fold cross-validation, and sequence-level predictions. Operates as an MCP server within Claude Code via containerized Docker deployment with automatic GPU mounting and host IPC for PyTorch shared memory. Supports multiple ProtTrans architectures (ProtT5-XL, ProtAlbert) with flexible CPU/GPU inference for varying memory constraints.