Install SecuRedact MCP
SecuRedact MCP is an open-source (Apache-2.0) local privacy layer for AI workflows. It detects and sanitizes PII locally before it reaches Claude, Gemini, or any external AI. It is distributed on PyPI and declared in the official MCP registry as io.github.GigantesHJI/securedact-mcp.
1. Install from PyPI
Requires Python 3.12 (tested on 3.12, below 3.13):
py -3.12 -m pip install "securedact-mcp[ml]"
Then run the setup wizard to enable the enforced Claude Code or Gemini CLI integration:
securedact-mcp setup
Or, if you are not installing a model:
py -3.12 -m pip install securedact-mcp securedact-mcp setup
2. Model consent
Detection models are downloaded separately and only with your explicit consent. The package never downloads at startup and does not ship model weights. Each model install requires roughly 2.4 GiB of disk space. You can choose English and/or Dutch detection models during setup.
3. Enforced integrations
-
Claude Code — via
securedact-mcp setup. The plugin intercepts prompts locally before the model runs and blocks or sanitizes sensitive content. Shipped, config-validated, and unit-tested. -
Gemini CLI — via
securedact-mcp setup. The extension runs the privacy check before the agent, the model, and external tools are used. - MCP-compatible clients — Codex, Cursor, and Windsurf work in MCP mode (no enforced hooks). See the repository docs for setup templates.
4. Verify your installation
Confirm the server is available:
securedact-mcp
For a complete walkthrough of configuration, policies, and MCP client setup, see the official documentation.