MCP / privacy proxy / open source

Your AI sees aliases.
Your secrets stay yours.

OpenMaskit is an MCP proxy that strips sensitive values from tool responses before they reach your AI agent — and puts them back when the agent calls home.

§ 01

Every MCP tool response your agent reads becomes part of its context window. Hostnames, tokens, customer emails, internal IPs — once they're in the prompt, they're somewhere: in transit, in logs, in someone else's training pipeline. The agent doesn't need the real values to do its job. It just needs something to refer to them by.

§ 02
CLIENT AI agent Claude · Cursor · Codex · …
mcp / http
PROXY OpenMaskit localhost · open source
stdio / http
UPSTREAM MCP server Slack · GitHub · Postgres · …

Drop OpenMaskit between your agent and the MCP servers it already talks to. Responses pass through; sensitive values are replaced with stable aliases on the way out, and swapped back to real values on the way in.

§ 03

Everything runs on your machine.

Masking happens locally. Aliases are computed locally. The audit log is local, encrypted at rest. Real values, masked values, tool-call payloads — none of it ever crosses the wire to us. Read the privacy notice →

§ 04
# requires Python 3.10+ (uvx fetches one if needed)
uvx openmaskit

Don't have uv? curl -LsSf https://astral.sh/uv/install.sh | sh. Then open http://127.0.0.1:9473 — the docs walk you through your first masking rule.