This server gives you two MCP tools, mask_text and unmask_text, that strip PII before it hits your cloud LLM and restore it afterward. Detection runs locally using regex for structured identifiers like SSNs, credit cards, PESEL, and IBAN, plus an optional Ollama model for names and organizations. It replaces detected values with opaque tokens like [PERSON:1] or [SSN:1] and holds the mapping in memory per session. Supports English and Polish out of the box with heuristic language detection. This is defense in depth, not anonymization. The masked structure and context still go to the LLM, and any missed entities travel in cleartext. Reach for this when you need to reduce direct identifier exposure in prompts while keeping your workflow intact.
claude mcp add --transport stdio woladi-pseudonym-mcp uvx pseudonym-mcp