SecretCarousel
The secret vault coding agents use - store, rotate, and share credentials via CLI or MCP. Claim tokens let one agent hand a secret to another, one-time and scoped, even across tenants; every access is AES-256-GCM encrypted and audit-logged.
README
SecretCarousel
The secret vault your coding agents run themselves — store, rotate, and hand off credentials from any agent, so they stop passing keys in plain text.
git for your code. sc for your agents' secrets.
Your agent touches credentials all day — a database URL here, an API key there — and the moment its context resets, those secrets end up pasted into a .env file in Slack, echoed into a log, or left in the transcript. You become the vault. SecretCarousel is the vault the agent runs itself: one command, and it stores secrets encrypted, rotates them on a schedule, and — the part nothing else does — hands a secret to another agent through a one-time, scoped, expiring claim token, even across companies. No plain-text value ever crosses the boundary.
Works with: Claude Code · Cursor · Cline · Windsurf · Aider · Codex · any MCP client
Install
npm install -g secretcarousel
The npm package is secretcarousel; the command is sc.
Or self-provision a vault in one command — no email, no card:
npx secretcarousel signup my-project --local
Quick Start
# Self-provision a tenant (API key saved to .sc/config.json)
sc signup my-project --local
# Store a secret — AES-256-GCM encrypted at rest; returns an id, never echoes the value
sc secret "DATABASE_URL" "postgres://user:pass@host/db" -t database-credentials
# Retrieve by id — decrypted on demand, access-logged
sc secret show sec-abc123
# List / search — ids + names, never values
sc secrets
sc search stripe
# Rotate now
sc rotate sec-abc123
# Full reference
sc --help
Claim tokens — how agents hand off secrets
This is the part nothing else does. One agent mints a one-time, scoped, expiring token; another agent — different project, even different company — redeems it exactly once; the token burns; the whole exchange is audit-logged. The plain-text value never crosses the boundary as a copy-pasteable string.
# Agent A: hand a secret to another tenant/agent
sc claim "<the-secret-value>" --to partner-tenant --contract CTR-ID
# Agent B: redeem the token you were handed (once, audited)
sc claim redeem CLM-TOKEN
# List the tokens you've minted + their status
sc claims
No more DMing a key and hoping the other side deletes it. The token is single-use, time-boxed, and revocable, and every mint/redeem lands in the audit trail.
Zero-knowledge mode — even we can't read it
By default, secrets are encrypted at rest with AES-256-GCM. Flip on zero-knowledge mode and encryption happens on your machine, before the value ever reaches the server — SecretCarousel stores only ciphertext and never holds the key to unlock it. A database breach leaks gibberish.
# Store a zero-knowledge secret (encrypted client-side)
sc secret "SIGNING_KEY" "$(cat key.pem)" --e2e
# Retrieve — auto-decrypts locally with your device key
sc secret show sec-def456
Multiple agents can share one zero-knowledge vault via an enrollment handshake — the vault key is passed device-to-device, never to the server:
sc enroll init # first device bootstraps the vault
sc enroll create --minutes 15 # mint an enrollment token for another agent
sc enroll redeem <token> # joining agent unlocks the same vault
sc enroll status # vault + local device state
Rotation — set it and forget it
# Rotate immediately
sc rotate sec-abc123
# Schedule rotation with alerts
sc rotate set sec-abc123 --schedule "30d" --email ops@myproject.dev
Sharing — time- and view-limited links for humans
# One-hour, single-view link a teammate can open in a browser
sc share sec-abc123 --hours 1
# List active shares
sc shares
Environments — .env in, .env out
# Export a project's secrets as a .env file
sc env pull --env production > .env
# Promote secrets from one environment to another
sc env promote --from staging --to production
MCP Server
Prefer tools over a CLI? sc ships an MCP server. Point Claude Code (or any MCP client) at it and your agent gets the full toolset — store, retrieve, list, search, rotate, share, and the claim-token handoff — with structured JSON responses.
claude mcp add secretcarousel -- sc mcp-serve
For clients that use a JSON config (Cline, Cursor, Windsurf), pass your API key via the SC_API_KEY environment variable. The MCP server runs outside your project directory, so it will not pick up .sc/config.json:
{
"mcpServers": {
"secretcarousel": {
"command": "sc",
"args": ["mcp-serve"],
"env": { "SC_API_KEY": "sc_your_key_here" }
}
}
}
No key yet? Start it without one — the server boots in onboarding mode with an sc_signup tool that provisions a vault instantly (48-hour activation window; the emailed $0 card check starts the 7-day free trial), then the full toolset unlocks in place.
Remote MCP — zero install
No CLI at all? Claude Web, Claude Desktop, Raycast, or any hosted MCP client can connect straight to our remote server. Same toolset, same API key, nothing to install:
URL: https://mcp.secretcarousel.com/sse
Auth: Authorization: Bearer sc_YOUR_API_KEY
Connect without the header and it starts in onboarding mode — call sc_signup to self-provision a vault mid-session (48-hour activation window, 7-day free trial via emailed $0 card check). Setup guide: Remote MCP quickstart.
Audit trail — provable, exportable
Every store, retrieve, rotate, share, mint, and redeem lands in an immutable event log. Read it from the terminal or export it for compliance.
sc audit --limit 50
sc audit export --format csv > audit.csv
Features
- Secrets — store, retrieve, update, delete; AES-256-GCM at rest; typed (database-credentials, api-key, certificate, and more)
- Zero-knowledge mode — client-side encryption; the server only ever sees ciphertext, multi-device vaults via an enrollment handshake
- Claim tokens — hand a secret to another agent/tenant via a one-time, scoped, expiring, audited token
- Rotation — rotate now or on a schedule, with email alerts
- Sharing — time- and view-limited links for humans
- Environments — export as
.env, promote secrets between environments - Scoped API keys — read/write/admin per key, project-scoped
- MCP server — local (
sc mcp-serve) or fully remote (mcp.secretcarousel.com): Claude Code, Claude Web, Cursor, Raycast, any MCP client - Audit trail — immutable event log, exportable as CSV or JSON
- Agent self-signup — an agent provisions its own vault mid-session, no human in the loop
Pricing: 7-day free trial. Card required (secure Stripe checkout), nothing charged during the trial — cancel before it ends and you pay nothing. After that, from $9/mo: Starter $9 · Pro $29 · Scale $99, each with included secrets plus low overage. Unlimited agents on every plan, no per-seat pricing. Details.
Per-Project Config
sc login --local --key YOUR_KEY # saves to .sc/config.json (project-local)
sc me # show tenant info + plan + usage
Config is per-project and auto-loaded from .sc/config.json. Add .sc/ to your .gitignore.
Agent Integration
Add to your CLAUDE.md, .cursorrules, .clinerules, .windsurfrules, or AGENTS.md:
## SecretCarousel
This project uses SecretCarousel as its secret vault.
Use the `sc` CLI. Config is in .sc/config.json (auto-loaded).
If not configured: sc login --local --key YOUR_KEY
Never paste a secret value into code, logs, or chat — store it with
`sc secret "NAME" "VALUE"` and retrieve by id at the moment of use.
To hand a secret to another agent, mint a claim token — never send the raw value.
Why this exists
Every coding agent handles credentials, and every one of them eventually leaks one — into a .env committed by mistake, a log line, a Slack paste, a transcript that outlives the session. I wanted the agent to own its secrets the way it owns its code: a vault it runs itself, where the value is fetched only at the moment of use and handed off without ever becoming a copy-pasteable string. It's early and iterating fast — if something's rough or missing, open an issue.
Documentation
License
Proprietary — Tyga.Cloud Ltd. 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。