Scan text and files for accidentally-committed secrets (AWS keys, GitHub tokens, Stripe keys, RSA keys) via @mukundakatta/secretsniff-mcp (npx)

@mukundakatta/secretsniff-mcp v0.1.1 — verified recipe

Package: @mukundakatta/secretsniff-mcp (npm) Transport: stdio Launch: npx @mukundakatta/secretsniff-mcp (binary: mcp-secretsniff) Auth: none required Tools: 2 — scan_text, scan_file

Tool schemas

`scan_text` — scan a string for secrets

• text (string, required): text to scan
• min_entropy (number, default 4.5): Shannon entropy threshold for high-entropy fallback
• min_entropy_length (integer, default 32): minimum substring length for entropy rule
• include_high_entropy (boolean, default true): toggle the high-entropy catch-all rule

`scan_file` — read a file from disk and scan it (same return shape + file path)

• path (string, required): absolute or relative file path
• Same optional params as scan_text

Tested capabilities

6/6 calls succeeded, 100% pass rate, p50=8ms

Detected secret types

From the tested corpus, the scanner reliably detects:

• AWS_ACCESS_KEY — AKIA… pattern
• SLACK_TOKEN — xoxb-… / xoxp-… / xoxa-… pattern
• STRIPE_KEY — sklive… / sktest… pattern
• RSA_PRIVATE_KEY — -----BEGIN RSA PRIVATE KEY----- marker
• GENERIC_API_KEY — variable assignment patterns like API_KEY = "…"
• HIGH_ENTROPY — catch-all for strings with Shannon entropy ≥ 4.5 bits/char and length ≥ 32

Return shape

{
  "count": 3,
  "findings": [
    {
      "kind": "AWS_ACCESS_KEY",
      "line": 1,
      "column": 14,
      "start": 13,
      "end": 33,
      "matched": "AKIAIOSFODNN7EXAMPLE",
      "entropy": 3.684
    }
  ]
}

Key observations

• Fast: median call takes ~8ms; first call is slower (~261ms) due to regex compilation
• Line+column positions: findings include exact line, column, and byte offsets — ready for IDE integration or CI pipeline annotations
• Shannon entropy scoring: every finding includes entropy in bits/char, useful for ranking severity
• Clean code returns empty: {count:0, findings:[]} — no false positives on normal code
• `scan_file` resolves symlinks: on macOS /tmp → /private/tmp, path in result shows the resolved path

Gotchas

1. GitHub tokens (`ghp_…`) are detected as HIGH_ENTROPY, not a specific GITHUB_TOKEN kind — the scanner uses entropy rather than a dedicated regex for these
2. Database connection strings (e.g. postgres://user:password@host/db) are NOT detected — no pattern for embedded credentials in URIs
3. SendGrid keys (SG.xxx…) were NOT detected in testing — may need a newer version or custom rule
4. `scan_file` path must exist on disk — the server reads the file itself, not through the agent
5. First call is ~20× slower than subsequent calls (~261ms vs ~8ms) — likely regex compilation overhead