privacyscrubber-mcp
Related Servers
Alternatives to privacyscrubber-mcp
- AlicenseBqualityBmaintenanceEnables local, in-memory de-identification of PII, PHI, and credentials with deterministic reversible surrogate tokens before LLM transmission, then restores originals from volatile session mappings and audits for exposures without network egress or disk writes.51Apache 2.0
Related Servers
- AlicenseNot gradedqualityBmaintenanceEnables local, in-memory sanitization of sensitive text before LLM ingestion by replacing PII, PHI, and credentials with deterministic reversible tokens, then restoring the originals in responses. It also audits content for exposed secrets and PII and provides compliance receipts without external network egress or disk writes.7 npm1Apache 2.0
- AlicenseAqualityDmaintenanceScans prompts for PII and masks or redacts sensitive data locally before sending to an LLM, supporting multiple anonymization modes.1MIT

scrubguardofficial
AlicenseAqualityCmaintenanceEnables AI assistants to scrub or reversibly tokenize PII and secrets in text and nested JSON before sending prompts to LLMs, then restore the original values from a vault map afterward. Runs locally with no network calls and exposes scrub_text, rehydrate_text, and scrub_json tools.340 npmMIT- AlicenseNot gradedqualityCmaintenanceStrips personal data and credentials from text before it reaches a model, log, or ticket, using checksum-verified detection for cards, IBANs, routing numbers, SSNs, contact details, and API keys. Placeholders such as <EMAIL_1> can later be swapped back in so the model's response is restored with the original values.MIT
- FlicenseAqualityBmaintenancePII sanitization layer for autonomous AI agent pipelines. Detects and redacts emails, phone numbers, national IDs, private keys, and financial data before text reaches LLMs. Supports EN, ES (LATAM), PT (BR/PT), DE, JA.12-
- AlicenseNot gradedqualityDmaintenanceLocal-first CLI and MCP server for redacting sensitive text before sharing logs, configs, and errors with AI tools.MIT
TDQS
Scored across 28 tools
The surface contains near-duplicate tool families: sanitize_text vs ztds_sanitize, reveal_text vs ztds_restore, audit_context vs ztds_audit, and guard_session_reset vs ztds_reset_session all perform essentially the same operations under different branding. File handling also overlaps heavily across sanitize_file, guard_read_file, and redact_file, and status reporting is split between check_status, ztds_info, and guard_session_info. Descriptions help somewhat (STEP 0/1/3 guidance) but an agent still faces real misselection risk.
All names are snake_case, but the set mixes three conventions: bare verb_noun (sanitize_text, export_audit_log), ztds_-prefixed (ztds_sanitize, ztds_audit), and guard_-prefixed (guard_exec, guard_read_file), with some inconsistent hybrids like check_status and stream_reveal_text. The prefixes appear to encode product tiers/categories rather than actions, which is readable but not a predictable verb_noun pattern.
28 tools is heavy for a sanitize/restore domain, and a meaningful share are redundant aliases (ZTDS variants duplicating the core sanitize/reveal/audit trio, plus two session-reset tools). The set could be roughly halved without losing capability.
The lifecycle is well covered: pre-flight audit, sanitize (text/file/RAG), restore (batch and streaming), in-place redaction, directory scanning, session reset, config scaffolding, agent rules, and compliance/audit export. Minor gaps exist, such as no tool to inspect or list active token mappings and no streaming sanitize counterpart to stream_reveal_text.