Ama

Posted on May 31

🚀 I Built Trade MCP: Remote MCP Server for Crypto Tools and Safer AI Trading Workflows

#mcp #ai #development #programming

I built Trade MCP.

GitHub: https://github.com/AmaLS367/TradeMcp

And no, this is not another “AI trading bot that will make you rich while you sleep” thing.

That stuff sounds cool until you remember one small detail:

letting an AI agent freely touch your exchange account is insane.

So I built Trade MCP around a different idea:

AI should research, compare and prepare.
Humans should approve dangerous actions.

Simple. Safer. Actually usable.

🧠 What is Trade MCP?

Trade MCP is a remote MCP server and dashboard for crypto workflows.

It connects AI clients to:

exchange data
market data providers
encrypted API key storage
Earn comparison tools
portfolio context
human-approved trading workflows
a dashboard for managing everything

The goal is not to replace your brain.

The goal is to give your AI assistant real tools without turning it into a financial gremlin with API keys.

⚠️ The problem

MCP is powerful.

But when crypto gets involved, things become dangerous very quickly.

A normal AI crypto setup usually looks like one of these:

ChatGPT with no live data
A local script with messy .env files
An agent with too much access
A trading bot that can act before you even understand what happened

That is not good enough.

If money is involved, the system needs:

encrypted credentials
auth
policies
logs
dashboard visibility
human approval
clean tool boundaries

So I started building that.

🔥 What Trade MCP supports

🔐 Encrypted exchange connections

Trade MCP is built for exchange connections like:

Binance
Bybit

API keys are not supposed to just sit in random config files.

They are encrypted with AES-256-GCM before storage.

Because “just put your exchange key in plain text” is not a security strategy. It is a cry for help.

📊 Market data providers

AI models do not magically know live market data.

So Trade MCP supports external provider connections for market context.

Examples:

CoinGecko
CryptoPanic
TAAPI.IO
Twelve Data
OANDA
NewsAPI
Messari

This lets AI answer using actual data instead of confident nonsense.

And confident nonsense in trading is expensive.

💰 Earn comparison

Crypto Earn products are annoying to compare manually.

Different platforms.
Different APY.
Different assets.
Different lock periods.
Different conditions.

Trade MCP is built to help aggregate and compare those opportunities so an AI client can give a more useful answer.

Not just:

“Maybe staking is good.”

But actual structured comparison.

🌐 Remote MCP endpoint

Local MCP is cool for small scripts.

But for a real tool, remote MCP feels much better.

Trade MCP gives you a remote MCP server approach:

one server
one endpoint
centralized auth
centralized tool management
dashboard-based setup
easier deployment
cleaner multi-client usage

This makes the project feel less like a hacky local experiment and more like infrastructure.

🖥️ Dashboard

Trade MCP also has a React dashboard.

Because editing everything through config files gets old fast.

The dashboard is meant for:

managing exchange connections
managing providers
checking configuration
controlling workflows
making the server usable by humans, not only terminal goblins

A tool can be powerful, but if the UX is painful, people will not use it.

🧩 Tech stack

The project uses:

TypeScript
Node.js
Express
React
Vite
Firebase Auth
Firestore
CCXT
MCP SDK
Docker
Vitest
Zod

So it is not just a lonely script in a folder named final-final-real-version.

It has a real app structure, backend, frontend, docs, tests and deployment setup.

🛡️ My philosophy for AI trading tools

I do not think AI agents should instantly execute trades with full freedom.

That sounds impressive in a demo.

In real life, it is how you speedrun regret.

A better workflow is:

🤖 AI should:

research
compare data
check market context
summarize balances
find Earn opportunities
prepare structured proposals
explain possible risks

👤 Human should:

review
approve
reject
adjust
decide

That is the core idea behind Trade MCP.

Let AI prepare.
Let humans approve.

⚙️ Example workflow

Imagine this:

You connect Binance or Bybit
You add market data providers
You connect your AI client to Trade MCP
You ask something like:

Compare my USDT Earn options and find better available opportunities.

Or:

Analyze BTC market context and prepare a possible trade plan, but do not execute anything.

The AI can use MCP tools to gather context.

Then it can prepare an answer or proposal.

You stay in control.

That is the important part.

🧨 Why this is useful

Because AI agents are becoming more capable.

But more capability without more control is not automatically good.

Especially in crypto.

Trade MCP tries to give the agent useful abilities while still keeping the dangerous parts behind proper structure.

That means:

no blind autopilot trading
no raw key chaos
no random local config soup
no “trust the model bro” energy

Just a cleaner foundation for AI-assisted crypto workflows.

🚧 Current status

Trade MCP is still evolving.

Things I want to improve:

better AI client connection guides
more provider integrations
stronger policy profiles
better trade proposal flow
clearer risk summaries
audit logs
dashboard polish
more tests
better examples

So no, this is not “finished forever”.

But it is already a serious foundation.

✅ TL;DR