Waqi: an MCP proxy that redacts PII from tool responses before the model sees them
AggressiveAnxiety481 · reddit · 2026-09-09
A builder shipped Waqi, a hosted MCP proxy addressing a real gap: when MCP servers connect to tools like Stripe, Xero, or CRMs, tool responses flow verbatim into the model's context — customer emails, card metadata, and phone numbers land in a third-party context window.
The pipeline is detect → redact/pseudonymise → log, applied mid-flight to every response:
- Structured detection: Luhn-gated cards, mod-97 IBANs, emails, phones, gov IDs, secrets, IPs — checksum gating avoids shredding order references that merely look like card numbers
- Deterministic per-value placeholders ([EMAIL:f4b1]) so the model can still reason that "the same customer failed 3 times" without seeing who
- Secrets are hard-redacted, never pseudonymised
- Per-member MCP URLs and per-call audit rows
Building the browser demo surfaced a real bug: print-format IBANs ("GB29 NWBK 6016 1331 9268 19") half-matched as phone numbers, fixed with spaced-group matching plus checksum validation on the compacted string. It's a paid product with a free in-browser demo; local NER for names-in-prose is next.
More from coding & agent
- Plausible Analytics MCP server lets AI assistants query website traffic stats — modelcontextprotocol · 2026-09-09
- Human Design MCP connector brings bodygraph chart analysis to AI assistants — modelcontextprotocol · 2026-09-09
- One change cut Claude Code tokens 3x: 10.4M→3.7M, $9.21→$2.81 with InsForge — Roger_M_Taylor · 2026-09-09
- Google releases 1-hour graph engineering tutorial: from single agent to self-improving 24/7 systems — Roger_M_Taylor · 2026-09-09
- Personal AI Workflows Are the New Company Assets: Workers May Leave With Their AI Agents — aigclink · 2026-09-09
- Agent given a computer builds its own simulation-with-agents, in a simulation — Confident_Salt_8108 · 2026-09-09