Armor Crypto MCP — agentic threat model

Not certain from the listing — relies on the host application's underlying LLM (e.g., Claude via Claude Desktop). Threats include prompt injection leading to unauthorized transaction generation or signature requests.

Not certain from the listing — does not explicitly mention RAG or vector databases, but handles blockchain state data. Threats include poisoning of price feeds or transaction history data.

Implements the Model Context Protocol (MCP) to expose tools for wallet management, swaps, and staking. Vulnerabilities include tool misuse, where malicious prompts trick the agent into executing unauthorized transfers or interacting with malicious smart contracts.

Deployed as an MCP server, typically running locally or in a container. Key custody is the primary infrastructure risk; if the host environment is compromised, private keys or API credentials used for transaction signing can be exfiltrated.

Not certain from the listing — no built-in logging, guardrails, or transaction simulation features are described. Lack of observability could allow silent transaction manipulation or unauthorized fund movement to go unnoticed.

Crucial layer for this agent as it handles financial transactions. The listing notes that 'key custody and action confirmation are paramount,' but does not detail built-in multi-sig, hardware wallet integration, or mandatory human-in-the-loop (HITL) confirmation.

As an MCP tool, it operates within a broader ecosystem where other agents or LLMs can invoke it. A compromised orchestrator agent could abuse this tool to drain wallets or execute unfavorable trades without user consent.