keyblind

Keyblind - Blind AI to Your Keys

Encrypted secrets vault with MCP for AI agents. Secrets are resolved at runtime, never leaked to LLM conversations.

Why

100,000+ LLM conversations with exposed secrets were found indexed by search engines. AI agents read .env files, copy-paste secrets into chat, and commit them accidentally. Keyblind keeps secrets encrypted at rest (AES-256-GCM SQLite vault) and resolves them just-in-time over the Model Context Protocol - the plaintext value never touches the transcript.

Install

npm i -g keyblind      # install
keyblind init          # create your encrypted vault
keyblind setup-mcp     # auto-configure MCP for Claude Code

For other editors, add config-template.json to your MCP config:

{
  "mcpServers": {
    "keyblind": {
      "command": "keyblind",
      "args": ["start"]
    }
  }
}

Then restart your agent.

Use

keyblind set MY_API_KEY      # store encrypted (prompts securely)
keyblind sandbox                 # agents see deterministic fakes in .env
keyblind get MY_API_KEY      # resolve the decrypted value
keyblind run -- npm start        # run a command with secrets as env vars
keyblind list                    # names only - values never shown

The agent calls MCP tools such as resolve_secret, store_secret, list_secrets, totp_code, share_secret, and more (16 tools total).

Features

• 16 MCP tools - secrets CRUD, TOTP 2FA, encrypted sharing, dead man's switch, .env sandbox, audit log
• Works everywhere - Claude Code, Cursor, Copilot, Windsurf, Cline, Zed
• Zero network, zero telemetry - fully local, no cloud, no accounts
• Machine-identity-bound key - encryption key wrapped with a machine fingerprint

Pricing

Free: 5 secrets with all 16 tools. Fully usable - no purchase required.