secretsniff-mcp
MukundaKatta/secretsniff-mcp
MCP server: scan text/code for accidentally-committed secrets. AWS/GitHub/Slack/Stripe keys, JWTs, RSA keys, high-entropy strings.
Instalação
npx convez-ai MukundaKatta/secretsniff-mcp1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
README
A tradução para o português deste README ainda está na fila.
secretsniff-mcp
An MCP server that gives AI assistants the ability to scan text or code for accidentally-committed secrets before they ship.
Works with Claude Desktop, Cursor, Cline, Windsurf, Zed, and any other MCP client.
What it catches
| Detector | Pattern |
|---|---|
AWS_ACCESS_KEY | AKIA… 20-char keys |
GITHUB_TOKEN | ghp_/gho_/ghu_/ghr_/ghs_… |
SLACK_TOKEN | xox[baprs]-… |
STRIPE_KEY | sk_live_/sk_test_/pk_live_/pk_test_… |
JWT | three url-safe base64 segments |
RSA_PRIVATE_KEY | -----BEGIN RSA PRIVATE KEY----- marker |
SSH_PRIVATE_KEY | -----BEGIN OPENSSH PRIVATE KEY----- marker |
GENERIC_API_KEY | api[_-]?key\s*[=:]\s*"…" style assignments |
HIGH_ENTROPY | strings ≥ 32 chars with Shannon entropy ≥ 4.5 bits/char |
Tools exposed
scan_text
Scan a string for secrets. Returns one entry per finding with line/column, the matched substring, and Shannon entropy.
{ "text": "const KEY = 'AKIAIOSFODNN7EXAMPLE';" }
→
{
"findings": [
{
"kind": "AWS_ACCESS_KEY",
"line": 1,
"column": 14,
"start": 13,
"end": 33,
"matched": "AKIAIOSFODNN7EXAMPLE",
"entropy": 3.78
}
]
}
scan_file
Read a file from disk and scan it. Same return shape as scan_text plus
the file path. Useful when an agent has just written a file and wants to
double-check it before committing.
Configure your MCP client
Claude Desktop's claude_desktop_config.json:
{
"mcpServers": {
"secretsniff": {
"command": "npx",
"args": ["-y", "@mukundakatta/secretsniff-mcp"]
}
}
}
Cursor / Cline / Windsurf / Zed configs follow the same command + args
pattern.
Why this is not a --lint flag
This is meant for the agentic loop: an LLM that just wrote a config or
edited a file calls scan_text on its own output before declaring success.
Existing CLI scanners (gitleaks, trufflehog) target the
human-driven pre-commit workflow; they're great at that and I'd still run
them in CI. secretsniff-mcp covers the gap during the LLM's editing turn.
License
MIT.