Keyway MCP Server

Let AI manage your secrets securely

Keyway is a GitHub-native secrets manager. This MCP server lets AI assistants like Claude securely access your secrets without ever exposing them in conversation.

Installation · Tools · Security · Development

Why Keyway MCP?

Traditional secret management with AI is risky: copying secrets into chat exposes them in logs and context. Keyway MCP solves this:

| Without Keyway | With Keyway MCP | |----------------|-----------------| | Copy secrets into chat | Secrets stay in vault | | Visible in conversation history | Never exposed to AI | | Manual secret creation | Generate securely, never exposed | | Hope AI doesn't leak them | Cryptographically protected |

Key features:

• Zero exposure — Generate, validate, and use secrets without the AI ever seeing them
• Pre-deployment validation — Check all required secrets exist before shipping
• Secret scanning — Detect leaked credentials in your codebase
• Environment diffing — Compare secrets across dev/staging/prod

Quick Install

Prerequisites

First, authenticate with Keyway CLI:

npx @keywaysh/cli login

Claude Code

claude mcp add keyway -- npx @keywaysh/mcp

VS Code / Cursor

code --add-mcp '{"name":"keyway","command":"npx","args":["-y","@keywaysh/mcp"]}'

Or click: Install in VS Code

Other IDEs

Add to your MCP config:

{
  "mcpServers": {
    "keyway": {
      "command": "npx",
      "args": ["-y", "@keywaysh/mcp"]
    }
  }
}

Settings → AI → Manage MCP Servers → Add:

{
  "mcpServers": {
    "keyway": {
      "command": "npx",
      "args": ["-y", "@keywaysh/mcp"]
    }
  }
}

/mcp add

Then enter npx -y @keywaysh/mcp when prompted.

Advanced settings → Extensions → Add custom extension

Select STDIO type, command: npx -y @keywaysh/mcp

Available Tools

keyway_generate

Generate secure secrets and store them directly in the vault. The value is never exposed to the AI.

"Generate a new JWT secret for production"

{
  "name": "JWT_SECRET",
  "type": "jwt-secret",
  "environment": "production"
}

Types: password | uuid | api-key | jwt-secret | hex | base64

Response:

{
  "success": true,
  "action": "created",
  "name": "JWT_SECRET",
  "type": "jwt-secret",
  "length": 43,
  "preview": "eyJh**********************************MDkz",
  "message": "Secret created. The actual value was never exposed in this conversation."
}

keyway_validate

Validate required secrets exist before deployment. Supports auto-detection from code.

"Check if production has all required secrets"

{
  "environment": "production",
  "required": ["DATABASE_URL", "STRIPE_SECRET_KEY", "JWT_SECRET"]
}

Or auto-detect from your codebase:

{
  "environment": "production",
  "autoDetect": true
}

Response:

{
  "valid": false,
  "missing": ["STRIPE_SECRET_KEY"],
  "present": ["DATABASE_URL", "JWT_SECRET"],
  "stats": {
    "requiredCount": 3,
    "presentCount": 2,
    "coverage": "66.7%"
  },
  "message": "✗ Missing 1 required secret in production: STRIPE_SECRET_KEY"
}

keyway_scan

Scan your codebase for leaked secrets. Detects 18+ secret types.

"Scan the codebase for leaked credentials"

{
  "path": "./src"
}

Detects: AWS keys, GitHub tokens, Stripe keys, Slack webhooks, private keys, and more.

Response:

{
  "filesScanned": 142,
  "findingsCount": 2,
  "findings": [
    {
      "file": "src/config.ts",
      "line": 23,
      "type": "GitHub PAT",
      "preview": "ghp_********************************xyz"
    }
  ]
}

keyway_diff

Compare secrets between environments.

"What's different between staging and production?"

{
  "env1": "staging",
  "env2": "production"
}

Response:

{
  "onlyInEnv1": ["DEBUG_MODE"],
  "onlyInEnv2": ["REDIS_CLUSTER_URL"],
  "different": [
    {
      "key": "DATABASE_URL",
      "preview1": "**st (45 chars)",
      "preview2": "**db (52 chars)"
    }
  ],
  "same": ["API_KEY", "JWT_SECRET"],
  "stats": {
    "totalEnv1": 10,
    "totalEnv2": 11,
    "different": 1
  }
}

keyway_inject_run

Run commands with secrets injected as environment variables.

"Run the test suite with production secrets"

{
  "command": "npm",
  "args": ["test"],
  "environment": "production"
}

Secrets are injected into the command's environment and masked in any output.

keyway_list_secrets

List secret names (not values) in an environment.

{
  "environment": "production"
}

keyway_set_secret

Create or update a secret manually.

{
  "name": "WEBHOOK_URL",
  "value": "https://hooks.example.com/abc123",
  "environment": "production"
}

keyway_list_environments

List available environments for the repository.

Security

Keyway MCP is designed with security as the primary concern:

| Feature | How it works | |---------|--------------| | Token encryption | Uses AES-256-GCM, same as Keyway CLI | | No secret logging | Values never appear in logs or output | | Output masking | inject_run redacts secrets from stdout/stderr | | Shell injection prevention | Commands run with shell: false | | File permissions | Validates ~/.keyway/.key is 0600 | | Generate, don't expose | keyway_generate creates secrets without revealing them |

What the AI can see

| Tool | AI sees value? | |------|----------------| | keyway_generate | No — only masked preview | | keyway_validate | No — only key names | | keyway_scan | No — only masked previews | | keyway_diff | No — only masked previews | | keyway_inject_run | No — values masked in output | | keyway_list_secrets | No — only key names | | keyway_set_secret | Yes — value provided by user |

Development

# Install dependencies
pnpm install

# Run in development
pnpm dev

# Build
pnpm build

# Run tests
pnpm test

# Lint & format
pnpm lint
pnpm format

Environment Variables

| Variable | Description | |----------|-------------| | KEYWAY_API_URL | Override API URL (default: https://api.keyway.sh) |

License

MIT — see LICENSE

keyway.sh · Built for developers who care about security