secret-scanner

@guardbee/mcp-secret-scanner

🇬🇧 English | 🇹🇷 Türkçe | 🇨🇳 中文 | 🇯🇵 日本語 | 🇰🇷 한국어 | 🇧🇷 Português | 🇪🇸 Español | 🇩🇪 Deutsch

An MCP server that scans your source files, directories, and environment configs for exposed API keys, passwords, tokens, and other secrets. Ask Claude directly whether your project is leaking secrets.

Telemetry is off by default. With GUARDBEE_TELEMETRY=1 it sends the server, tool, success, duration and short parameter values such as a file path; longer strings and fields like content, token or password are replaced with a placeholder. See @guardbee/mcp-telemetry.

---

Features

• 76 Secret Patterns — AWS, Azure, GCP, GitHub, GitLab, Stripe, OpenAI, Anthropic, OpenRouter, Groq, Hugging Face, Slack, Twilio, SendGrid, DigitalOcean, Cloudflare, Supabase, Vault, PyPI, Shopify, Linear, Notion, Atlassian and more — plus the Turkish providers iyzico, PayTR, Netgsm and İleti Merkezi, matched on their variable names since their keys have no prefix. The rules live in @guardbee/guard-core, so @guardbee/mcp-security-proxy masks the same formats in tool results at runtime
• File & Directory Scanning — a single file or an entire project tree
• Git History & Staged Scanning — --history scans every line any commit added, so a key deleted from the files but still in the repository is found, reported once at the commit that added it (with author and date); --staged scans only what you are about to commit
• Baseline — --write-baseline records today's findings as hashes (never the secrets, safe to commit); --baseline then reports only new ones, so an existing repo can adopt the scanner in CI without fixing everything first. SARIF results carry a stable partialFingerprints value
• High-Entropy Values — a random-looking value assigned to a secret-like name (WEBHOOK_SIGNING_SECRET=… in .env, client_secret: in YAML, "internalApiKey" in JSON) is reported even when no provider rule knows its format. Randomness is measured as Shannon entropy (hex values need 32+ characters; values without digits need more bits, so a camel-case passphrase does not count); public/publishable, *_hash, *_id and *_url names, placeholders and ${VAR} references are left alone, and a value a provider rule already reported is not reported twice. Medium severity, low in test files; --no-entropy (or entropy: false in the MCP tools) turns it off
• Agent Transcripts — scan --agent-history (MCP: scan_agent_history) scans what coding agents keep on your machine: Claude Code (~/.claude/projects, ~/.claude.json), Codex (~/.codex), Gemini CLI (~/.gemini/tmp) and Continue (~/.continue/sessions). A key pasted into a chat or printed by a command the agent ran stays there — and was already sent to the model provider, so rotate it. Each secret is reported once per agent with how often it appears; the entropy check is off here unless --entropy, since transcripts are full of code the agent read
• Large Line Files — .jsonl, .ndjson, .ipynb and .log files over 1 MB (transcripts, notebooks with outputs, logs) are read line by line instead of being skipped
• Live Verification (CLI only) — --verify asks each provider whether a found key still works, through a read-only identity endpoint (GitHub /user, OpenAI /models, Slack auth.test, …): GitHub, GitLab, OpenAI, Anthropic, Slack, Stripe, Hugging Face, npm, SendGrid, Groq, OpenRouter and DigitalOcean. Each key goes only to its own provider, once, and never appears in the output; findings are marked live / invalid / unknown and live ones are listed first. --fail-on=live fails CI only for keys that still work. The MCP tools never verify, so an agent cannot send your secrets anywhere on its own
• Smart Skipping — directories like node_modules, .git, dist, build, .next are skipped automatically
• Safe Redaction — matches are shown as first 4 + stars + last 4 characters
• Allowlist Support — allowlist known test/fake values
• 71 Unit Tests — 100% passing test suite

---

Quick Start

npm install -g @guardbee/mcp-secret-scanner

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "guardbee-secret-scanner": {
      "command": "npx",
      "args": ["-y", "@guardbee/mcp-secret-scanner"]
    }
  }
}

---

MCP Tools

Tool	Description
scan_text	Scans the given text for secrets
scan_file	Scans a single file
scan_directory	Recursively scans a directory and its subdirectories
scan_git	Scans a git repository's staged changes or every commit's added lines (history)
scan_agent_history	Scans coding agents' transcripts on this machine (Claude Code, Codex, Gemini CLI, Continue)
list_patterns	Lists all active secret patterns

Example Usage

You can ask Claude:

"Scan my project for secrets: /Users/me/my-app"

"Are there any secrets in this .env file?"

"Is this text safe: export API_KEY=sk-abc123..."

---

Detected Secret Types

Category	Examples
Cloud	AWS Access Key, AWS Secret, GCP API Key
Source Control	GitHub PAT, GitLab Token
Payments	Stripe Secret/Publishable Key
AI	OpenAI API Key, Anthropic API Key, HuggingFace Token
Communication	Slack Bot Token, Twilio Auth Token, SendGrid Key
Database	PostgreSQL URL, MySQL URL, MongoDB URI, Redis URL
Cryptography	RSA Private Key, EC Private Key, OpenSSH Key, PGP Key
Web	JWT Token, Bearer Token
Package/Platform	npm Token, Docker Hub Token, Vercel Token
Generic	SECRET=, PASSWORD=, API_KEY= patterns

---

Security Note

This tool partially redacts matched values in scan results (e.g. sk_live_abc1...xyz9). Full values are never logged or exported.

---

CLI — CI/CD Integration

In addition to MCP server mode, this can also be used directly as a CLI:

# Scan a project directory
npx @guardbee/mcp-secret-scanner scan ./my-project

# Scan a single file
npx @guardbee/mcp-secret-scanner scan .env

# Fail only on critical/high
npx @guardbee/mcp-secret-scanner scan . --fail-on=high

# JSON output (for CI reporting)
npx @guardbee/mcp-secret-scanner scan . --format=json
# Secrets pasted into Claude Code / Codex / Gemini CLI / Continue sessions on this machine
npx @guardbee/mcp-secret-scanner scan --agent-history
# Which of the found keys still work? (sends each key to its own provider)
npx @guardbee/mcp-secret-scanner scan . --history --verify --fail-on=live
# Pre-commit: only the lines you are about to commit
npx @guardbee/mcp-secret-scanner scan --staged

# Every commit on every branch (finds keys deleted from the files)
npx @guardbee/mcp-secret-scanner scan . --history
npx @guardbee/mcp-secret-scanner scan . --history=main..HEAD   # just this branch

# Adopt an existing repo: record known findings once, then fail only on new ones
npx @guardbee/mcp-secret-scanner scan . --history --write-baseline=.guardbee-secrets-baseline.json
npx @guardbee/mcp-secret-scanner scan . --history --baseline=.guardbee-secrets-baseline.json

Exit codes: 0 = no secrets found · 1 = secrets found · 2 = error

GitHub Actions

name: Secret Scan
on: [push, pull_request]

jobs:
  secret-scan:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 0   # full history for --history
      - name: Scan for exposed secrets
        run: npx @guardbee/mcp-secret-scanner scan . --history --baseline=.guardbee-secrets-baseline.json --fail-on=high

GitLab CI

secret-scan:
  image: node:20
  script:
    - npx @guardbee/mcp-secret-scanner scan . --fail-on=high
  only:
    - merge_requests
    - main

Pre-commit Hook

# .git/hooks/pre-commit
npx @guardbee/mcp-secret-scanner scan --staged --fail-on=high || exit 1

---

Development

npm install
npm test          # 71 unit tests
npm run build     # TypeScript compile

---

License

MIT — GuardBee

Feedback

Questions, ideas, or a false positive? Start a thread in GitHub Discussions or open an issue. If GuardBee is useful to you, a ⭐ on the repository helps others find it.