local-security

local-security

A Claude Code plugin that audits the security of your local Claude Code environment.

If you use Claude Code with MCP servers, OAuth tokens, API keys, or third-party plugins, this plugin helps you understand what's exposed and whether it's properly protected.

Install

Run these commands inside Claude Code:

/plugin marketplace add simplefy-ai/local-security
/plugin install local-security@simplefy-ai

Then restart Claude Code.

How it works

Two skills with different jobs:

Skill	What it does	When to run
/local-security:local-setup	Scans your environment and creates a security assessment document	Once, or when your setup changes significantly
/local-security:local-review	Runs automated checks and flags issues	Monthly, or after adding servers/credentials/permissions

Setup discovers what you have — MCP servers, credential files, privacy settings, data flows — and writes everything to ~/.claude/security-assessment.md. It documents but does not judge. Previous assessments are archived with timestamps.

Review validates what setup found. It checks:

1. Disk encryption (FileVault / LUKS)
2. MCP server health — are local servers clean, up to date?
3. Dependency vulnerabilities — npm audit / pip audit
4. Credential security — file permissions and gitignore coverage
5. Permissions audit — write tools at global scope, secrets embedded in permission entries, stale one-off bash commands piling up
6. Remote server version pinning — unpinned packages are a supply chain risk
7. Stale credentials — tokens that haven't refreshed in 90+ days
8. Anthropic data flow — telemetry, error reporting, privacy settings
9. Assessment status — is the security assessment file current?

Each check reports PASS, WARN, or FAIL. Review also populates a mitigation roadmap in the assessment file with prioritised actions for any issues found.

What risks does this cover?

Three things that actually matter for a local Claude Code setup:

1. Supply chain — Is the code running with your credentials auditable and trusted? Unaudited MCP servers pulled from PyPI or npm could intercept tokens or log data.
2. Credential storage — Are your OAuth tokens, API keys, and SSH keys properly protected on disk? File permissions and gitignore coverage matter.
3. Scope creep — Do your MCP tools have broader access than they need? Write tools at global scope mean Claude can modify external systems from any project.

Requirements

• Claude Code (CLI)
• macOS or Linux
• Git (for local MCP server integrity checks)
• npm or pip (for dependency auditing — optional)

Licence

MIT — see LICENSE.

Built by simplefy.ai