mintmcp/agent-security

Security plugins for Claude Code and Cursor

This repository currently provides a secrets scanner plugin.

Motivation

Coding agents are powerful, but we've repeatedly seen them read and propagate sensitive data during everyday work. That can be acceptable for casual "vibe coding" experiments, but it's not acceptable for production software engineering. We built this to make accidental leakage much harder: a standalone, local-first scanner with minimal footprint (no external dependencies, regex-only), running as editor/agent hooks entirely on your machine, and easy to set up so teams can adopt it without friction.

Install

Claude Code Plugin Marketplace

Install via the Claude Code plugin marketplace:

/plugin marketplace add mintmcp/agent-security
/plugin install secrets-scanner@agent-security

PyPI (manual hooks)

pipx install claude-secret-scan
# or
python3 -m pip install --user claude-secret-scan

Add hooks to ~/.claude/settings.json if using PyPI:

{
  "hooks": {
    "UserPromptSubmit": [
      {"hooks": [{"type": "command", "command": "claude-secret-scan --mode=pre"}]} 
    ],
    "PreToolUse": [
      {"matcher": "Read|read", "hooks": [{"type": "command", "command": "claude-secret-scan --mode=pre"}]}
    ],
    "PostToolUse": [
      {"matcher": "Read|read", "hooks": [{"type": "command", "command": "claude-secret-scan --mode=post"}]},
      {"matcher": "Bash|bash", "hooks": [{"type": "command", "command": "claude-secret-scan --mode=post"}]}
    ]
  }
}

Cursor

Copy examples/configs/cursor-hooks.json to ~/.cursor/hooks.json or configure similarly:

{
  "version": 1,
  "hooks": {
    "beforeReadFile": [{"command": "cursor-secret-scan --mode=pre"}],
    "beforeSubmitPrompt": [{"command": "cursor-secret-scan --mode=pre"}]
  }
}

Repository Layout

.
├── .claude-plugin/
│   └── marketplace.json
├── plugins/
│   └── secrets_scanner/
│       ├── .claude-plugin/
│       │   └── plugin.json
│       ├── hooks/
│       │   ├── hooks.json
│       │   └── secrets_scanner_hook.py
│       ├── tests/
│       │   └── read_hook_test.py
│       ├── TESTING.md
│       └── README.md
├── examples/
│   └── configs/
├── pyproject.toml
└── README.md

Notes

• Pre hooks block when secrets are detected. Post hooks print warnings.
• No external dependencies. The scanner uses only Python's built-in regexes and simple pattern matching.
• Runs completely locally. No code, prompts, or files leave your system.
• Curious how it works? See plugins/secrets_scanner/hooks/secrets_scanner_hook.py for the core implementation and patterns.
• No telemetry by default. For org-wide visibility and governance (dashboards, metrics, alerting), see https://mintmcp.com.

License

Apache License 2.0. See LICENSE.

Acknowledgements

Regex patterns were informed by or adapted from detect-secrets (Apache 2.0).

{
  "$schema": "https://anthropic.com/claude-code/marketplace.schema.json",
  "name": "agent-security",
  "version": "0.1.14",
  "description": "Security plugins for Claude Code and Cursor: secret scanning and safe coding patterns",
  "owner": {
    "name": "MintMCP",
    "email": "support@mintmcp.com"
  },
  "plugins": [
    {
      "name": "secrets-scanner",
      "description": "Scans for common credentials across cloud, source control, payment, and collaboration providers",
      "version": "0.1.14",
      "author": {
        "name": "MintMCP",
        "email": "support@mintmcp.com"
      },
      "source": "./plugins/secrets_scanner",
      "category": "security",
      "keywords": [
        "security",
        "secrets",
        "credentials",
        "scanning"
      ]
    }
  ]
}