@securecode/mcp-server

@securecode/mcp-server

MCP Server for SecureCodeHQ. Lets Claude Code access your secrets securely — without ever seeing them.

Zero-Knowledge AI Access

By default, secret values are never shown to the AI. When Claude reads a secret, the value is written to a local file on your machine. The AI gets the file path but never sees the actual value.

You: "Get my Stripe API key"
Claude: ✓ STRIPE_KEY injected → ~/.securecode/.session/a1b2c3d4.env

The value is NOT in this conversation.

This is inject mode — the default. If you explicitly need the AI to see the value, use reveal: true.

Setup (2 minutes)

1. Get your API key from securecodehq.com > Settings > API Keys

2. Add to your Claude Code config:

Edit ~/.claude.json:

{
  "mcpServers": {
    "securecode": {
      "command": "npx",
      "args": ["@securecode/mcp-server"],
      "env": {
        "SECURECODE_API_KEY": "sc_your_key_here"
      }
    }
  }
}

3. Restart Claude Code. That's it.

What Claude Can Do

Once connected, Claude Code can:

"Get my Stripe API key"              -> injects to local file (AI never sees value)
"Get my DB_URL with reveal"          -> reveals value to AI (audited)
"List my production secrets"         -> shows names and tags, never values
"Save this API key as OPENAI_KEY"    -> creates a new secret
"Import my .env.production file"     -> bulk import with auto-tagging
"Lock my session"                    -> blocks all access until you wake it
"Set up SecureCode"                  -> guided onboarding: signup, import, configure
"byebye"                            -> locks session + cleans secrets from disk

Available Tools

Tool	Description
get-secret	Get a secret — inject to file by default, reveal: true to show to AI, cleanup: true to remove files
list-secrets	List all secrets with tags, expiry status, and descriptions
create-secret	Create a new secret with tags, TTL, and domain
update-secret	Update value, description, tags, or domain
delete-secret	Soft-delete a secret
renew-secret	Renew expired secrets or change TTL
import-env	Import from .env or CSV content (with preview mode)
export-env	Export secrets as .env or CSV format
get-status	Check plan, usage limits, secrets count, and MCP server version
wake-session	Unlock session with optional tag scope and auto-sleep timer
sleep-session	Lock session immediately, block all access, clean injected files
session-status	Check session status, scope, and time remaining
byebye	End session: lock + clean all injected secrets from disk + goodbye
get-active-rules	List active MCP access rules (read-only)
onboard	Guided setup: sign up, import .env, auto-create API key, configure MCP — all from Claude Code
help	Get SecureCode documentation: tools, SDK setup, sessions, rules, troubleshooting

Guided Onboarding

New to SecureCode? Just tell Claude:

You: "Set up SecureCode for this project"
Claude: [calls onboard]

Step 1: Open this link to create your account → (opens signup popup)
Step 2: Open this link to import your .env   → (opens import popup)
Step 3: API key created and MCP configured    → (Claude does it automatically)

The entire setup takes under 3 minutes without leaving your terminal:

• Your .env files stay untouched — SecureCode is for your AI agent to access secrets without reading .env directly.
• The API key is auto-created and configured — it never appears in the conversation if you let the agent do it.
• After restart, all SecureCode tools work immediately.

MCP Access Rules

Control how AI agents access your secrets with tag-based policies. Rules are created from the dashboard and enforced server-side.

Action	Effect
Block Always	Secret is only accessible from the dashboard
Require Confirmation	Agent must acknowledge before accessing
Require Session	Requires an active session (wake-session)
Block Models	Only allows specific AI models
Notify	Sends email notification on access (non-blocking)

You: "Read my STRIPE_LIVE_KEY"
Claude: [MCP Rule: "Block production secrets"]
        Access blocked. This secret is only accessible from the dashboard.

Rules are evaluated by priority: Block Always > Require Session > Block Models > Require Confirmation > Notify.

Auto-Update Check

The MCP server checks for updates on first use. If your version is outdated, you'll see a warning with every tool response:

⚠ Your SecureCode MCP server is outdated (v0.6.1 → v0.7.0).
To update, restart Claude Code — it will fetch the latest version automatically if you use npx.

Security Tips (Tip Engine)

The MCP server includes an adaptive Tip Engine that educates developers about security best practices:

• First session: Welcome message with best practices (shown before first secret)
• Production access: Warning when accessing production-tagged secrets (shown before secret)
• Stale secrets: Reminder to rotate credentials older than 90 days
• Generic names: Suggestion to use descriptive secret names
• No TTL: Recommendation to set expiration on frequently-used secrets

Tips are throttled (max 3 per session, 1-week cooldown per secret) so they inform without annoying.

Session Lock

Control access to your secrets per session:

You: "Wake my session for the acme project staging"
Claude: [calls wake-session with scope [{project:"acme", env:"staging"}]]
       -> Session unlocked. Only acme/staging secrets accessible.

You: "Lock my session"
Claude: [calls sleep-session]
       -> Session locked. All access blocked. Injected files cleaned.

You: "byebye"
Claude: -> Session locked & secrets cleaned from disk. See you next time!

Sessions auto-sleep after configurable inactivity (default: 2 hours).

How It Works

• Zero-knowledge by default: Secret values are written to a local file — the AI never sees them
• Reveal mode: Explicit reveal: true returns value to AI (audited as conscious action)
• Auto-cleanup: Injected files are removed on sleep, byebye, process exit (SIGINT/SIGTERM), or next inject
• Parallel sessions: Multiple Claude Code instances don't collide (file hash based on API key + PID)
• Secrets are encrypted with AES-256-GCM using envelope encryption (Cloud KMS)
• Every access is logged with AI model, IP, machine identity, access mode (inject/reveal), and timestamp
• API keys are scoped to your account with sc_ prefix
• The MCP server runs locally via stdio transport — secrets never pass through third parties
• Device approval required on first use from each machine
• MCP Access Rules enforce tag-based policies server-side before secret delivery

Requirements

• Node.js >= 18
• A SecureCodeHQ account with an API key

License

MIT