zKettle
Self-hosted, zero-knowledge encrypted, self-destructing secrets for secure agent-to-agent coordination
README
zKettle
Self-hosted zero-knowledge expiring secrets. Encrypt locally, store ciphertext on the server, share a URL with the decryption key in the fragment. The server never sees the plaintext or the key.
Installation
Go install (requires Go 1.25+)
go install github.com/benderterminal/zkettle@latest
This installs to $GOPATH/bin (typically ~/go/bin). Make sure it's in your PATH: export PATH="$HOME/go/bin:$PATH"
Binary download
curl -fsSL https://github.com/benderterminal/zkettle/releases/latest/download/zkettle-$(uname -s | tr A-Z a-z)-$(uname -m | sed 's/x86_64/amd64/;s/aarch64/arm64/') -o /usr/local/bin/zkettle && chmod +x /usr/local/bin/zkettle
From source
git clone https://github.com/benderterminal/zkettle.git && cd zkettle && make install
Via AI agent
Paste this prompt into Claude Code, Cursor, or any MCP-compatible agent:
I want to set up zKettle — a self-hosted zero-knowledge secret sharing tool. Install it with go install github.com/benderterminal/zkettle@latest, then read the MCP setup instructions in the README at https://github.com/benderterminal/zkettle. When configuring the MCP server, use the absolute path to the installed binary (find it with which zkettle or check ~/go/bin/). Once configured, test the full workflow using the CLI: create a secret, read it back, and revoke it. Note that MCP servers are loaded at startup — the new tools won't be available until the next terminal session.
Quick Start
# Start the server with a Cloudflare tunnel (instant public URL)
zkettle serve --tunnel
# Or start locally
zkettle serve --port 3000
# Create a secret (in another terminal)
echo "my secret password" | zkettle create --views 1 --minutes 60
# → http://localhost:3000/s/abc123#key
# Read a secret
zkettle read "http://localhost:3000/s/abc123#key"
# → my secret password (use -c for clipboard, -o <file> for file output)
# Revoke a secret
zkettle revoke --server http://localhost:3000 --token <delete-token> <id>
Open the URL in a browser to reveal the secret via the web viewer.
Docker Deployment
# Build and run with Docker Compose
docker compose up -d
# Or build and run manually
docker build -t zkettle .
docker run -d -p 3000:3000 -v zkettle-data:/data zkettle
The container listens on port 3000 and stores data in /data. Configure with environment variables (see Configuration Reference).
Production Deployment
With TLS (direct)
zkettle serve --host 0.0.0.0 --tls-cert /path/to/cert.pem --tls-key /path/to/key.pem
With a reverse proxy (recommended)
Run zkettle behind Caddy, Nginx, or Traefik for automatic TLS:
zkettle serve --host 127.0.0.1 --trust-proxy
Enable --trust-proxy so zkettle reads the real client IP from X-Forwarded-For headers.
Systemd
Copy the service template and enable it:
sudo cp contrib/zkettle.service /etc/systemd/system/
sudo systemctl daemon-reload
sudo systemctl enable --now zkettle
The service template uses DynamicUser=yes with ReadWritePaths=/var/lib/zkettle, so systemd manages the data directory automatically.
Configure via environment file at /etc/zkettle/env:
ZKETTLE_PORT=3000
ZKETTLE_HOST=0.0.0.0
ZKETTLE_ADMIN_TOKEN=your-secret-token
ZKETTLE_TRUST_PROXY=true
Backups
The database is a single SQLite file at <data-dir>/zkettle.db. Back it up with:
sqlite3 /var/lib/zkettle/zkettle.db ".backup /backups/zkettle-$(date +%Y%m%d).db"
Admin API
Enable the admin endpoint by setting an admin token via environment variable:
export ZKETTLE_ADMIN_TOKEN=my-secret-admin-token
zkettle serve
Note: Passing
--admin-tokenon the command line exposes the token in process listings (ps,/proc/*/cmdline). Prefer the environment variable or config file.
List active secrets
# Via CLI
zkettle list --server http://localhost:3000 --admin-token my-secret-admin-token
# Via API
curl -H "Authorization: Bearer my-secret-admin-token" http://localhost:3000/api/admin/secrets
Returns metadata only (ID, views remaining, timestamps). No encrypted content or decryption keys are ever exposed.
GET /api/admin/secrets
Returns 404 when no admin token is configured (endpoint disabled). Requires Authorization: Bearer <token> header.
Response (200):
[
{
"id": "abc123...",
"views_left": 2,
"expires_at": "2024-01-02T03:04:05Z",
"created_at": "2024-01-01T00:00:00Z"
}
]
Metrics
Enable the /metrics endpoint with the --metrics flag:
export ZKETTLE_ADMIN_TOKEN=my-secret-admin-token
zkettle serve --metrics
The /metrics endpoint requires the admin token (Authorization: Bearer <token> header). Returns 404 when no admin token is configured.
Returns JSON metrics at GET /metrics:
{
"zkettle_secrets_active": 5
}
MCP Setup
zKettle includes an MCP server for use with Claude Desktop, Claude Code, or any MCP-compatible agent.
Important: Use the absolute path to the
zkettlebinary. Many MCP clients do not inherit your shell's PATH, so a barezkettlecommand will silently fail to start.
Add to your MCP client's configuration file:
{
"mcpServers": {
"zkettle": {
"command": "/absolute/path/to/zkettle",
"args": ["mcp", "--port", "3001", "--tunnel"]
}
}
}
Use --tunnel for public shareable URLs via Cloudflare Quick Tunnel (no account required). Omit it for local-only access. Use --base-url https://your-domain.com if you have a custom domain.
Claude Code shortcut:
claude mcp add -s user zkettle -- /absolute/path/to/zkettle mcp --port 3001 --tunnel
The MCP server starts an HTTP backend on the specified port and communicates with the agent over stdio. All encryption and decryption happens locally — in the browser (Web Crypto API), CLI, or MCP server process. The zKettle HTTP server never sees plaintext.
Available tools:
| Tool | Description |
|---|---|
create_secret |
Encrypt and store a secret (content or file input), returns an expiring URL |
read_secret |
Retrieve and decrypt a secret (file or clipboard output to avoid context exposure) |
list_secrets |
List active secrets (metadata only) |
revoke_secret |
Delete a secret by ID |
generate_secret |
Generate a random secret, optionally store it (create=true) |
CLI Reference
zkettle serve [options] Start the HTTP server
--port 3000 HTTP port
--host 127.0.0.1 Listen address (use 0.0.0.0 for all interfaces)
--data ./data Data directory for SQLite database
--base-url "" Base URL for generated links (default: http://localhost:{port})
--cors-origins "" Comma-separated allowed CORS origins
--tunnel Expose server via Cloudflare Quick Tunnel
--trust-proxy Trust X-Forwarded-For headers (behind a reverse proxy)
--log-format "" Log format: json or text (defaults to text)
--tls-cert "" TLS certificate file path
--tls-key "" TLS private key file path
--admin-token "" Admin API bearer token (enables GET /api/admin/secrets)
--max-secret-size 0 Max encrypted secret size in bytes (0 = 500KB)
--metrics Enable /metrics endpoint
zkettle create [options] Encrypt and store a secret (reads from stdin)
--views 1 Max views before auto-delete
--minutes 1440 Minutes until expiry (default 24h)
--server http://localhost:3000 Server URL
--json Output JSON to stdout
--quiet, -q Suppress stderr output
zkettle read [options] <url> Retrieve and decrypt a secret (quote the URL)
--clipboard, -c Copy to clipboard instead of printing to stdout
--file, -o <path> Write to file (0600 permissions) instead of stdout
zkettle revoke [options] <id> Delete a secret
--server http://localhost:3000 Server URL
--token "" Delete token (returned by create, or set ZKETTLE_DELETE_TOKEN)
zkettle list [options] List active secrets (requires admin token)
--server http://localhost:3000 Server URL
--admin-token "" Admin API bearer token
--json Output raw JSON
zkettle generate [options] Generate a cryptographically random secret
--length 32 Length in characters
--charset alphanumeric Character set: alphanumeric, symbols, hex, base64url
zkettle mcp [options] Start MCP server on stdio with HTTP backend
--port 3000 HTTP port for API
--host 127.0.0.1 Listen address
--data ./data Data directory
--base-url "" Base URL for generated links
--tunnel Expose server via Cloudflare Quick Tunnel
--trust-proxy Trust X-Forwarded-For headers (behind a reverse proxy)
--log-format "" Log format: json or text (defaults to text)
zkettle version Print version
Configuration Reference
Configuration is resolved in order of precedence: flags > env vars > config file > defaults.
Config file
Search order: ./zkettle.toml, $HOME/.config/zkettle/zkettle.toml
port = 3000
host = "127.0.0.1"
data = "./data"
base_url = ""
cors_origins = []
trust_proxy = false
tunnel = false
log_format = "" # defaults to "text"
tls_cert = ""
tls_key = ""
admin_token = ""
max_secret_size = 0 # 0 = 500KB default
metrics = false
Security: If your config file contains
admin_token, restrict permissions:chmod 600 zkettle.toml
Environment variables
| Variable | Description |
|---|---|
ZKETTLE_PORT |
HTTP port |
ZKETTLE_HOST |
Listen address |
ZKETTLE_DATA |
Data directory |
ZKETTLE_BASE_URL |
Base URL for generated links |
ZKETTLE_CORS_ORIGINS |
Comma-separated CORS origins |
ZKETTLE_TRUST_PROXY |
Trust proxy headers (true/1/yes) |
ZKETTLE_TUNNEL |
Enable Cloudflare tunnel (true/1/yes) |
ZKETTLE_LOG_FORMAT |
Log format: json or text |
ZKETTLE_TLS_CERT |
TLS certificate file path |
ZKETTLE_TLS_KEY |
TLS private key file path |
ZKETTLE_ADMIN_TOKEN |
Admin API bearer token |
ZKETTLE_MAX_SECRET_SIZE |
Max encrypted secret size in bytes |
ZKETTLE_METRICS |
Enable metrics endpoint (true/1/yes) |
ZKETTLE_DELETE_TOKEN |
Delete token for zkettle revoke (alternative to --token) |
API Reference
POST /api/secrets
Create a secret.
Request:
{
"encrypted": "<base64url ciphertext>",
"iv": "<base64url 12-byte IV>",
"views": 1,
"minutes": 1440
}
Constraints:
encrypted— required, base64url-encoded, max 500KB decoded (configurable via--max-secret-size)iv— required, base64url-encoded, must decode to exactly 12 bytesviews— 1-100 (default: 1)minutes— 1-43200 (default: 1440)
Response (201):
{
"id": "abc123",
"expires_at": "2024-01-02T03:04:05Z",
"delete_token": "def456"
}
Errors: 400 (validation), 415 (wrong Content-Type), 429 (rate limited)
GET /api/secrets/{id}
Retrieve and consume a view. Returns the encrypted blob:
{
"encrypted": "<base64url ciphertext>",
"iv": "<base64url IV>"
}
Errors: 400 (invalid ID format), 404 (expired, consumed, or nonexistent)
GET /api/secrets/{id}/status
Check availability without consuming a view.
Response (200):
{
"status": "available"
}
Errors: 400 (invalid ID format), 404 (expired, consumed, or nonexistent)
DELETE /api/secrets/{id}
Delete a secret. Requires Authorization: Bearer {delete_token} header.
Response: 204 No Content
Errors: 400 (invalid ID format), 401 (missing token), 404 (not found or wrong token)
GET /health
Health check. Returns 200 with {"status":"ok"}, or 503 with {"status":"error"} if the database is unavailable.
GET /s/{id}
Serves the web viewer HTML. The decryption key is in the URL fragment (#key) and never sent to the server.
Security Model
- Zero-knowledge: The server stores only AES-256-GCM ciphertext. The decryption key lives in the URL fragment, which browsers never send to the server.
- Client-side encryption: All encryption and decryption happens locally — in the browser (Web Crypto API), CLI, or MCP server process. The zKettle HTTP server never sees plaintext.
- Expiring: Secrets auto-delete after the configured number of views or time limit.
- Composable library: When importing zKettle as a Go library, use
ExtraRoutesandMiddlewareto extend the server with custom routes and middleware for any deployment. - CLI/MCP plaintext exposure: When reading secrets via CLI (
zkettle read) or MCP (read_secret), the decrypted plaintext appears in terminal output or the MCP tool result (which enters the AI agent's conversation context). To keep secrets out of terminal scrollback and agent logs, use--clipboard/--file(CLI) or theclipboard/fileparameters (MCP). When creating secrets via MCP, use thefileparameter to read content from a file, orgenerate_secretwithcreate=trueto generate and encrypt without exposing plaintext. In the web viewer, use the "Copy Without Revealing" button to copy a secret to clipboard without rendering it in the page.
Building
make build # Build for current platform
make build-all # Build for darwin/linux/windows amd64 + darwin/linux arm64
make install # Build and install to $GOPATH/bin or /usr/local/bin
make test # Run all tests
make clean # Remove build artifacts
License
AGPL-3.0 — see LICENSE.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。