zero-trace-lite
A read-only local MCP server that scans plain-text content for supported leak patterns and reports them with masked values.
scan_path— scans a folder or a single text file and reports findings.scan_text— scans a text snippet passed inline (optionalnamelabel).Detects home paths containing a login name, common secret-key formats and private keys, inline passwords (
PGPASSWORD,MYSQL_PWD,REDISCLI_AUTH, client password options), database URIs with passwords, secrets in.claude/settings*.jsonand.claude/.credentials.json, and e-mail addresses.Reports only — never blocks, edits, deletes, or commits anything; values are masked and no network connection is made.
No findings means no supported pattern matched, not proof that the content is secret-free; binary formats (ZIP, Office, PDF, images) are not opened.

Zero-Trace Lite
Add the same badge to your own README once the check runs on your repository: copy the line above, or see https://zero-trace-ship-kit.zeus-lab-global.workers.dev/badge/?utm_source=github_repo .
A small local check before your work leaves your machine.
A free CLI and local MCP server for supported leak patterns in plain-text files—including secrets left in Claude Code settings.
Python 3.9+ · Standard library only · Local stdio MCP · MIT-licensed Python files
Try the browser scanner | Get the free download ($0+) | See the full kit ($39 / Team $149)
More from ZEUS LAB: every product, side by side · the studio · free pack ($0): included files and a sample record

Still image. The 15-second demo plays on the demo page.
Quick start
git clone https://github.com/ZEUSLABZERO/zero-trace-lite.git
cd zero-trace-lite
python3 zero_trace_lite.py path/to/folderIt exits with code 1 when anything is found. Values are masked in the report. The tool opens no network connection.
Clicked "Don't ask again" in Claude Code? Approved commands are saved as literal strings in .claude/settings.local.json, including any password or token typed inline. The file is normally excluded from Git, but adding an already tracked file to .gitignore does not untrack it.
Related MCP server: secretscan
What it reports (plain-text files)
Secrets inside the permission rules of
.claude/settings.local.jsonand.claude/settings.json, and the.claude/.credentials.jsonfileInline passwords: the
PGPASSWORD/MYSQL_PWD/REDISCLI_AUTHvariables, the password option of command-line clients, database URIs that carry a passwordCommon secret-key formats and private keys
Home paths with a login name (macOS, Linux, Windows)
E-mail addresses
Obvious placeholders such as <password> or $DB_PASS are not reported.
Local MCP server
mcp_server_lite.py lets Claude Desktop, Claude Code or Cursor ask for the same report over stdio.
{"mcpServers": {"zero-trace-lite": {"command": "python3", "args": ["/absolute/path/to/mcp_server_lite.py"]}}}Tools: scan_path (a folder or a text file) and scan_text (a snippet). Read-only. The server imports nothing outside the Python standard library. That removes third-party package drift; it does not remove supply-chain risk as a whole. Read the two files before you run them — they are short.
GitHub Action
- uses: actions/checkout@v4
- uses: ZEUSLABZERO/zero-trace-lite@main
with:
path: "." # optional
fail-on-findings: "true" # optional; "false" reports without failing the jobThe result is written to the job summary. The runner needs python3 (GitHub-hosted runners have it).
Lite / Solo / Team
Lite (this repository) | Solo $39 | Team $149 | |
Report on plain-text files, CLI | ● | ● | ● |
Local MCP server | ● (report only) | ● | ● |
Git hooks that stop the commit or push on supported findings | — | ● | ● |
Checks inside ZIPs (nested), photo GPS, PDF and Office author fields | — | ● | ● |
Cleaned distribution ZIPs, cost guard | — | ● | ● |
People covered | anyone (MIT) | 1 | up to 10 in one organization |
License | MIT (the two Python files here) | commercial, one-time | commercial, one-time |
For a setup that also relays reviews between AIs and shows the activity on a local board: Autonomous Harness ($149).
Scope and license
It reads plain text only. ZIP, Office, PDF and image files are not opened.
It reports. It does not stop a commit or push.
"No findings" means no supported pattern was detected; it does not prove that the files contain no secrets.
A credential that reached Git history or a remote must be revoked at its issuer. Deleting the file does not take it back.
git ls-files .claude # is it tracked?
git log --oneline --all -- .claude/settings.local.json # was it ever committed?
git rm --cached .claude/settings.local.json # stop tracking (the file stays on disk)License: zero_trace_lite.py, mcp_server_lite.py and action.yml are released under the MIT License (see LICENSE). The paid Zero-Trace Ship Kit is a separate product under its own commercial license and is not part of this repository. © 2026 ZEUS LAB
If this saved you a worry before a push, a star helps other people find it.
日本語の要約: 公開しようとしているフォルダの文章ファイルから、Claude Code の設定ファイルに残った秘密情報、インラインのパスワード、秘密キー、ユーザー名入りのパス、メールアドレスを一覧にする無料の道具です。手元で動き、外部へ接続しません。報告のみで、commit / push は止めません。検出0件は「対応パターンが見つからなかった」という結果であり、秘密情報が存在しないことの証明ではありません。この2つのPythonファイルと action.yml は MIT ライセンスです。commit / push を止めるGitフックなどを含む有料の Zero-Trace Ship Kit は、別の商用ライセンスの別商品です。
Available Tools
2 toolsscan_pathA
Report supported leak patterns in a folder or a text file. Reads plain text only: home paths with a login name, common secret-key formats, inline passwords, database URIs with a password, secrets in .claude/settings*.json, e-mail addresses. Read-only.
Inputs: path (required).
scan_textA
Report supported leak patterns in a piece of text. Reads plain text only: home paths with a login name, common secret-key formats, inline passwords, database URIs with a password, secrets in .claude/settings*.json, e-mail addresses. Read-only.
Inputs: text (required), name.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 2 tools
The two tools differ by input source: scan_path takes a folder or file, while scan_text takes an inline string. The descriptions make this boundary explicit, though the tool names could sound more distinct since both perform the same underlying scan.
Both names use the same verb_noun snake_case pattern and start with the same verb, scan, so they form a completely predictable pair.
A lite, read-only leak scanner can reasonably be this small, but with only two closely related tools the surface feels thin. It is unlikely that routing many inputs needs more than these two tools.
The scanner covers detection for paths and text, the two main input shapes, but the lifecycle stops at reporting. There is no way to redact or export findings, list or configure supported patterns, or scan non-text sources, which are visible gaps for a leak-scanning domain.
Maintenance
Related MCP Connectors
Scan configs, files, or text for leaked secrets and obvious misconfigurations. Nothing stored.
Scan text and git diffs for committed credentials using real vendor key formats plus entropy.
Checks AI-written code for leaked keys and made-up packages, on your machine. Never uploads code.
Find leaked API keys, tokens and passwords in pasted text, see which are live, get a clean copy.
Related MCP Servers
- AlicenseNot gradedqualityCmaintenanceScans text and files for common secrets (AWS, GitHub, etc.) and redacts them to prevent credential leakage in AI-assisted development. Runs entirely locally with no telemetry.MIT
- AlicenseAqualityDmaintenanceScans diffs, files, and snippets for leaked secrets like AWS keys, GitHub tokens, and private keys, returning redacted findings while running fully locally without network calls.1MIT

Skarnofficial
AlicenseNot gradedqualityBmaintenanceScans AI coding sessions and assistant configs for leaked secrets and risky hooks; local, redacted37MIT- AlicenseNot gradedqualityBmaintenanceEnables an agent to scan text for leaked credentials locally before it writes or commits it, flagging known key, token, private-key and sensitive-filename patterns with redacted, line-level findings and no network calls. Detection is deliberately specific to avoid false positives that would block unattended agents.MIT