Cloak MCP Server
Local-first secret sanitization intercepting sensitive credentials before LLM exposure.
Data last scanned today · Reviewed today
Overview
CloakMCP operates as a local-first middleware that automatically replaces sensitive strings—like API keys and passwords—with innocuous tags before your code reaches an LLM. It allows you to maintain full LLM utility during development while ensuring that your actual credentials never leave your local environment. The tool integrates with any LLM, including Claude, Gemini, and Ollama, by serving as a transparent proxy. For power users, the Claude Code integration automates this cycle, scrubbing secrets at the start of a session and restoring them upon completion without manual intervention.
Our verdict
This server carries a high-risk rating with a security score of 24/100. The primary concern is that it lacks a read-only mode and requires full write access to function, which creates a significant security surface area. Additionally, the project is maintained by a single contributor and has not been editorially reviewed, making it suitable only for experimental local environments rather than mission-critical infrastructure.
- Zero direct dependencies minimize the software supply-chain footprint.
- First-class integration support for Claude Code automated workflows.
- Comprehensive documentation provided via DeepWiki.
- High-risk configuration due to required full write access.
- Single-contributor maintenance increases bus-factor risk.
- Project has not undergone external editorial security review.
- Inactivity indicated by no commits for over 150 days.
Compatibility
| Client | Local | Docker | Remote | Read-only |
|---|---|---|---|---|
| Claude Desktop | ||||
| Cursor | ||||
| VS Code | ||||
| Windsurf | ||||
| ChatGPT |
Frequently asked questions
›What are the system requirements for running CloakMCP?
CloakMCP requires Python 3.10 or higher installed on your machine.
›Does CloakMCP require a cloud connection to function?
No, CloakMCP is a local-first application and does not have any cloud dependencies; all operations are performed directly on your machine.
›How does the automatic protection for Claude Code work?
CloakMCP uses hooks that automatically pack files when a Claude Code session starts and unpack them when the session ends, ensuring secrets are replaced with deterministic tags while the LLM is working.
›Can I use CloakMCP with LLMs other than Claude?
Yes, the core CLI is LLM-agnostic and can be used to manually pack and unpack code for any LLM, including local Ollama models, GitHub Copilot, or Gemini.
›How can I customize the detection rules for secrets?
You can configure detection rules through YAML files, which support regex, entropy checks, and definitions for specific data types like IPs and URLs.
›Are there different levels of installation profiles available?
Yes, you can run 'cloak install' for the default secrets-only profile, or 'cloak install --profile hardened' to enable additional protections like bash safety and read guards.
Badge
Maintain this server? Add the live badge to your README.