Agent says:  fill #email with {{email}}, fill #password with {{password}}, click Sign In
agent-auth:  resolves {{email}} and {{password}} from encrypted vault, injects into browser
Agent gets:  "Login completed." (never sees the real values)

AI Agent                        agent-auth                       Browser
   |                               |                               |
   |  "Log into AWS"               |                               |
   |  steps: [                     |                               |
   |    fill #email {{email}}      |                               |
   |    fill #pass  {{password}}   |  1. Decrypt from local vault  |
   |    fill #totp  {{totp}}       |  2. Generate TOTP from seed   |
   |    click Submit               |  3. Inject via CDP  --------->| Form filled
   |  ]                            |  4. Zero memory               |
   |                               |                               |
   |  "Login completed" <--------- |                               |
   |                               |                               |
   |  (never saw any secrets)      |  (secrets wiped from RAM)     |

# Clone and install
git clone https://github.com/ex-nihilo-labs/agent-auth.git
cd agent-auth && bun install

# Create your vault (you'll set a passphrase)
bun run src/index.ts init

# Add a credential
bun run src/index.ts add github \
  --username "you@example.com" \
  --password "your-password" \
  --domains "github.com"

# Add one with TOTP
bun run src/index.ts add aws-root \
  --username "admin@company.com" \
  --password "hunter2" \
  --totp "JBSWY3DPEHPK3PXP" \
  --domains "signin.aws.amazon.com,console.aws.amazon.com"

# See what's stored (names only — no secrets shown)
bun run src/index.ts list
# → github (domains: github.com)
# → aws-root (domains: signin.aws.amazon.com, console.aws.amazon.com)

{
  "mcpServers": {
    "agent-auth": {
      "type": "stdio",
      "command": "bun",
      "args": ["run", "/path/to/agent-auth/src/index.ts", "serve"],
      "env": {
        "AGENT_AUTH_CDP_URL": "http://localhost:9222"
      }
    }
  }
}

bun run src/index.ts serve

{
  "service": "github",
  "url": "https://github.com/login",
  "steps": [
    { "action": "fill", "selector": "#login_field", "value": "{{email}}" },
    { "action": "fill", "selector": "#password", "value": "{{password}}" },
    { "action": "click", "selector": "input[type='submit']" },
    { "action": "wait", "selector": ".logged-in", "timeout": 5000 }
  ]
}

{
  "service": "openai",
  "url": "https://api.openai.com/v1/models",
  "method": "GET",
  "injection": "bearer"
}

agent-auth init                   # Create vault, set passphrase
agent-auth add <service>          # Add credential (interactive or with flags)
agent-auth list                   # List services (names only)
agent-auth remove <service>       # Delete a credential
agent-auth domains <service>      # View/edit allowed domains
agent-auth approve <code>         # Approve a pending auth request
agent-auth unlock                 # Unlock vault for current session
agent-auth lock                   # Lock vault, clear key from memory
agent-auth serve                  # Start MCP server (stdio)

agent-auth/
├── src/
│   ├── index.ts              # CLI dispatcher + MCP serve
│   ├── mcp/                  # MCP server (3 tools over stdio JSONRPC)
│   ├── browser/              # CDP injection via Playwright (fill/type/click/wait/select)
│   ├── placeholder/          # {{email}}, {{username}}, {{password}}, {{totp}} resolution
│   ├── vault/                # AES-256-GCM encrypted SQLite + Argon2id KDF
│   ├── totp/                 # TOTP generation via otpauth
│   ├── proxy/                # HTTP credential injection (6 methods)
│   ├── approval/             # Human approval gate + Pushover
│   ├── security/             # Rate limiter, input validator, domain allowlist
│   ├── audit/                # Append-only JSONL audit log
│   └── cli/                  # Human-only credential management
└── tests/                    # 65 tests across 6 suites

bun test              # 65 tests, 6 suites
bun run typecheck     # TypeScript strict mode

# Development
bun install
AGENT_AUTH_NO_KEYCHAIN=1 bun test    # Run tests
bun run typecheck                     # Type check