Skip to content

Security: InjectiveLabs/mcp-server

Security

SECURITY.md

Security Model

What the AI agent sees

  • Wallet addresses (inj1...)
  • Transaction hashes
  • Market data (prices, tickers)
  • Balances and positions

What the AI agent NEVER sees

  • Private keys
  • Raw key material
  • Mnemonics (after initial generation display)

Key storage

Private keys are encrypted at rest using AES-256-GCM with scrypt key derivation:

  • Random 32-byte salt per key file
  • Random 12-byte IV per key file
  • Directory: ~/.injective-agent/keys/
  • File permissions: 0o600 (owner read/write only)

Known trade-off: password exposure

Passwords are passed as function parameters and flow through MCP tool calls. If the MCP client logs tool invocations, passwords may appear in those logs.

Mitigation: Check your MCP client's logging configuration. Do not use high-value wallet passwords that are reused elsewhere.

Operational security

  • Use testnet for development and testing
  • Fund mainnet wallets with only as much as you're willing to lose
  • The library uses market orders — slippage is real, losses can exceed margin
  • Liquidation prices are estimates — actual liquidation depends on mark price oracles

There aren’t any published security advisories