AI · MCP · Automation
TradePulse
MCP-native trading agent with risk presets, news-driven execution and a full simulation mode — AI acts, but with guardrails.
4+
MCP tools for agents
3
Risk difficulty tiers
100%
Simulation rehearsal path
The challenge
Connecting LLM agents to real-world actions is easy to demo and hard to do safely. Trading needs strict risk limits, audit trails, secret isolation and a way to rehearse without real money.
What we built
TradePulse sits between three actors: LLM agents (MCP tools), an external news bot (JSON-RPC) and a human operator (cyberpunk dashboard). Every trade is sized by difficulty presets, clamped to position caps, and logged with a chain-of-thought line. Simulation mode runs the full pipeline offline.
Key capabilities
- MCP server exposing balance, market data, trade execution and news-trigger tools for LLM agents.
- HTTP JSON-RPC bridge for external news bots with optional Bearer token auth.
- Risk difficulty tiers (Easy / Medium / Hard) with leverage, position caps, SL/TP brackets.
- Cyberpunk React dashboard: balance, quick-strike trades, news ticker, system log.
- Full simulation mode with mock exchange and audit trail — zero exchange keys exposed to news bot.
Tech stack
Node.js · TypeScript · MCP SDK · ccxt · Express · React 19 · Vite · Zod · Tailwind v4