Trade MCP is a new open-source project that aims to solve one of the most obvious yet overlooked problems in AI-assisted crypto trading: why are we giving autonomous agents direct access to exchange accounts? Built by developer AmaLS367, Trade MCP positions itself as a remote Model Context Protocol server and dashboard for managing crypto workflows with an explicit human-approval layer baked into every dangerous action. The core philosophy is refreshingly straightforward: AI should research, compare data, and prepare structured proposals. Humans should review and approve anything that involves actual capital movement. The project explicitly rejects the "AI trading bot that makes you rich while you sleep" fantasy in favor of something more defensible. As AmaLS367 puts it: "A better workflow is: AI prepares. Humans approves." That framing immediately separates Trade MCP from the countless Telegram bots and sketchy Pine Script strategies flooding the crypto space.

Security Architecture

The security model starts with how API keys are handled. Rather than scattering exchange credentials across random .env files or config scripts, Trade MCP encrypts all API keys using AES-256-GCM before storage. Currently supported exchanges include Binance and Bybit, with what appears to be an extensible architecture for adding more. The project also integrates CCXT as the underlying trading library, giving it access to a wide range of exchange APIs through a unified interface. Beyond credential encryption, the system includes centralized authentication, policy controls, audit logging, and dashboard visibility over all tool boundaries. This is infrastructure thinking applied to AI agent permissions โ€” not just slapping a rate limit on a script and calling it done.

Market Data Integrations

Trade MCP connects to external market data providers so AI clients can operate with actual context rather than hallucinated price data. Supported integrations include CoinGecko, CryptoPanic, TAAPI.IO, Twelve Data, OANDA, NewsAPI, and Messari. The Earn comparison feature aggregates crypto staking and lending products across platforms โ€” handling the tedious variables like APY variance, lock periods, and asset eligibility that make manual comparison painful.

Remote MCP Server Design

Unlike local MCP setups suited for quick scripts, Trade MCP runs as a remote server with centralized endpoint management. This approach supports one server, one endpoint, shared authentication, unified tool management, and dashboard-based configuration โ€” making it practical to deploy once and connect multiple AI clients simultaneously without duplicating credentials or access logic across machines.

React Dashboard

A React-powered dashboard handles the operational side of things: managing exchange connections, configuring providers, monitoring workflows, and controlling what tools are available to connected AI clients. AmaLS367 acknowledges that powerful tools with terrible UX get abandoned, so the interface is treated as a first-class concern rather than an afterthought.

Tech Stack

The project uses TypeScript across both frontend and backend, Node.js with Express for the server layer, React with Vite for the dashboard, Firebase Auth and Firestore for identity and data storage, CCXT for exchange connectivity, MCP SDK for protocol compliance, Docker for deployment packaging, Vitest for testing, and Zod for schema validation. This is not a weekend hack cobbled together from tutorials โ€” it has tests, documentation, CI-ready structure, and an actual architecture diagram worth studying.

Current Status

The project is actively evolving. Planned improvements include better AI client connection guides, additional provider integrations, stronger policy profiles, more robust trade proposal flows, clearer risk summaries, audit log enhancements, dashboard refinements, expanded test coverage, and concrete usage examples. The creator acknowledges the work is not finished but frames it as a serious foundation rather than an MVP held together by hope.

Key Takeaways

  • API keys are encrypted with AES-256-GCM before storage โ€” no more plaintext credential sprawl
  • Human approval gate keeps AI agents in research-and-prepare mode, blocking unsupervised execution
  • Remote MCP server architecture enables centralized auth and multi-client deployments from one endpoint
  • Earn comparison aggregates crypto lending products across platforms into structured comparisons

The Bottom Line

Trade MCP is the kind of project that makes you wonder why it didn't exist sooner. As AI agents become more capable in financial contexts, handing them unrestricted exchange access is exactly backward โ€” more capability without better control is a liability, not a feature. This project takes a pragmatic stance: give your agent useful tools, but keep the dangerous ones behind human decision-making gates. That is the right call.