ccxt-mcp
CCXT MCP Server bridges the gap between AI models and cryptocurrency trading by providing a standardized interface through the Model Context Protocol. Created to empower automated trading strategies, this tool allows AI assistants like Claude and GPT to directly interact with over 100 cryptocurrency exchanges without requiring users to write comple
Documentation
CCXT MCP Server
ํ๊ตญ์ด ๋ฒ์ (Korean version)
CCXT MCP Server is a server that allows AI models to interact with cryptocurrency exchange APIs through the Model Context Protocol (MCP). This server uses the CCXT library to provide access to more than 100 cryptocurrency exchanges and their trading capabilities.
๐ Quick Start
# Install the package globally
npm install -g @lazydino/ccxt-mcp
# Run with default settings
ccxt-mcp
# or run without installation
npx @lazydino/ccxt-mcpInstallation and Usage
Global Installation
# Install the package globally
npm install -g @lazydino/ccxt-mcpRunning with npx
You can run it directly without installation:
# Using default settings
npx @lazydino/ccxt-mcp
# Using custom configuration file
npx @lazydino/ccxt-mcp --config /path/to/config.jsonView help:
npx @lazydino/ccxt-mcp --helpConfiguration
Registering the MCP Server in Claude Desktop
1. Open Claude Desktop Settings:
2. Add a New MCP Server:
3. Save and Test the Server:
Configuration Methods - Two Options
Option 1: Include Account Information Directly in Claude Desktop Settings (Basic Method)
This method includes CCXT account information directly in the Claude Desktop settings file (claude_desktop_config.json):
{
"mcpServers": {
"ccxt-mcp": {
"command": "npx",
"args": ["-y", "@lazydino/ccxt-mcp"],
"mcpBearerToken": "YOUR_MCP_TOKEN",
"accounts": [
{
"name": "bybit_main",
"exchangeId": "bybit",
"apiKey": "YOUR_API_KEY",
"secret": "YOUR_SECRET_KEY",
"defaultType": "spot"
},
{
"name": "bybit_futures",
"exchangeId": "bybit",
"apiKey": "YOUR_API_KEY",
"secret": "YOUR_SECRET_KEY",
"defaultType": "swap"
}
]
}
}
}Using this method, you don't need a separate configuration file. All settings are integrated into the Claude Desktop configuration file.
Option 2: Using a Separate Configuration File (Advanced Method)
To separate account information into a separate configuration file, set up as follows:
1. Create a Separate Configuration File (e.g., `ccxt-config.json`):
{
"mcpBearerToken": "YOUR_MCP_TOKEN",
"accounts": [
{
"name": "bybit_main",
"exchangeId": "bybit",
"apiKey": "YOUR_API_KEY",
"secret": "YOUR_SECRET_KEY",
"defaultType": "spot"
},
{
"name": "bybit_futures",
"exchangeId": "bybit",
"apiKey": "YOUR_API_KEY",
"secret": "YOUR_SECRET_KEY",
"defaultType": "swap"
}
]
}> Important: The configuration file must contain an `accounts` array at the root level, as shown above.
> Important: If you run the server in HTTP+SSE mode (`--sse`), set `mcpBearerToken` in the same config file. Clients must send `Authorization: Bearer ` on requests.
2. Specify the Configuration File Path in Claude Desktop Settings:
{
"mcpServers": {
"ccxt-mcp": {
"command": "npx",
"args": [
"-y",
"@lazydino/ccxt-mcp",
"--config",
"/path/to/ccxt-config.json"
]
}
}
}> Note: When using a separate configuration file with the `--config` option, the server will look for the `accounts` array directly in the root of the JSON file, not in `mcpServers.ccxt-mcp.accounts` path.
3. Running with External Configuration File from Command Line:
# Using custom configuration file
npx @lazydino/ccxt-mcp --config /path/to/ccxt-config.jsonYou can find an example configuration file at `config/ccxt-config.example.json` in the repository.
> Reasons to Use a Separate Configuration File:
>
> - Prevents recursive reference issues
> - Separates sensitive information like API keys
> - Easier multi-environment configuration (development, testing, production)
> - Improved configuration file version control
Key Features
- Market Information Retrieval:
- Trading Functions:
- Trading Analysis:
- Position Management:
- Risk Management:
- Technical indicator-based stop loss setting (e.g., lowest point among 10 candles on 5-minute chart)
- Volatility-based stop loss/take profit (ATR multiples)
- Maximum allowable loss limit (daily/weekly)
- Dynamic take profit setting (trailing profit)
How It Works
User AI Model(Claude/GPT) MCP Protocol CCXT MCP Server Cryptocurrency Exchange API1. User: Requests like "Tell me the Bitcoin price" or "Buy Ethereum on my Binance account"
2. AI Model: Understands user requests and determines which MCP tools/resources to use
3. MCP Protocol: Standardized communication between AI and CCXT MCP server
4. CCXT MCP Server: Communicates with cryptocurrency exchange APIs using the CCXT library
5. Exchange API: Provides actual data and executes trade orders
Using with AI Models
When registered with Claude Desktop, you can make the following types of requests to AI models:
Cautions and Recommended Prompts
When using AI models, consider the following cautions and use the prompt below for effective trading:
Your goal is to execute trades using the ccxt tools as much as possible
Cautions:
- Accurately identify whether it's a futures market or spot market before proceeding with trades
- If there's no instruction about percentage of capital or amount to use, always calculate and execute trades using the entire available capitalNotes:
- AI models sometimes confuse futures trading with spot trading.
- Without clear guidance on trading capital size, AI might get confused.
- Using the above prompt helps clearly communicate your trading intentions.
Basic Query Examples
Check and compare the current Bitcoin price on binance and coinbase.Advanced Trading Query Examples
Position Management
Open a long position on BTC/USDT futures market in my Bybit account (bybit_futures) with 5% of capital using 10x leverage.
Enter based on moving average crossover strategy and set stop loss at the lowest point among the 12 most recent 5-minute candles.Performance Analysis
Analyze my Binance account (bybit_main) trading records for the last 7 days and show me the win rate, average profit, and maximum consecutive losses.Detailed Trading Analytics
Analyze my trading performance on the bybit_futures account for BTC/USDT over the last 30 days. Calculate win rate, profit factor, and identify any patterns in my winning trades.Show me the monthly returns for my bybit_main account over the past 90 days and identify my best and worst trading months.Analyze my consecutive wins and losses on my bybit_futures account and tell me if I have any psychological patterns affecting my trading after losses.Development
Building from Source
# Clone repository
git clone https://github.com/lazy-dinosaur/ccxt-mcp.git
# Navigate to project directory
cd ccxt-mcp
# Install dependencies
npm install
# Build
npm run buildDocker
Build and Run with Docker Compose
1. Create a config file:
cp config/ccxt-config.example.json config/ccxt-config.jsonThen fill in your real API keys in `config/ccxt-config.json`.
Also set `mcpBearerToken` in that same config file.
2. Build image:
docker compose build3. Start the MCP server in background (SSE mode on localhost:2298):
docker compose up -d4. Verify local health endpoint:
curl -H "Authorization: Bearer YOUR_MCP_TOKEN" http://127.0.0.1:2298/healthzThis Compose setup runs MCP over HTTP+SSE (`/sse` + `/messages`) on localhost:2298 for reverse proxying.
Remote MCP Client Configuration
If your MCP client runs on another host, use an Nginx reverse proxy on this machine.
1. Copy `ccxt-mcp.nginx` to your Nginx config location and update certificate paths:
sudo cp ccxt-mcp.nginx /etc/nginx/conf.d/ccxt-mcp.conf2. Edit `/etc/nginx/conf.d/ccxt-mcp.conf` and set:
- `ssl_certificate`
- `ssl_certificate_key`
- Authorization header forwarding (already included in this file)
3. Reload Nginx:
sudo nginx -t && sudo systemctl reload nginx4. In your MCP client, configure the MCP server URL to your TLS endpoint:
- SSE URL: `https://YOUR_HOSTNAME_OR_IP:42299/sse`
- Messages URL: `https://YOUR_HOSTNAME_OR_IP:42299/messages`
- Header: `Authorization: Bearer YOUR_MCP_TOKEN`
You can generate a strong token with:
openssl rand -hex 32Config example:
{
"mcpBearerToken": "YOUR_MCP_TOKEN",
"accounts": [
{
"name": "bybit_main",
"exchangeId": "bybit",
"apiKey": "YOUR_API_KEY",
"secret": "YOUR_SECRET_KEY"
}
]
}Port note:
- `42298` is HTTP (redirect only)
- `42299` is HTTPS
- `2298` comes from `CCXT` on a phone keypad
If your MCP client accepts command-based MCP servers instead of URL configuration, configure `ccxt-mcp` with:
- Command: `docker`
- Args:
[
"run",
"--rm",
"-i",
"-p",
"127.0.0.1:2298:2298",
"-v",
"/absolute/path/to/ccxt-mcp/config/ccxt-config.json:/config/ccxt-config.json:ro",
"ccxt-mcp:local",
"--sse",
"--host",
"0.0.0.0",
"--port",
"2298",
"--config",
"/config/ccxt-config.json"
]Build the image first with `docker compose build`.
๐ค Contributing
Contributions are welcome! Please feel free to submit a Pull Request.
๐ License
Distributed under the MIT License. See the LICENSE file for more information.
โค๏ธ Support
If you find this project useful, please consider giving it a โญ๏ธ on GitHub!
Frequently asked questions
What is ccxt-mcp?
ccxt-mcp is CCXT MCP Server bridges the gap between AI models and cryptocurrency trading by providing a standardized interface through the Model Context Protocol. Created to empower automated trading strategies, this tool allows AI assistants like Claude and GPT to directly interact with over 100 cryptocurrency exchanges without requiring users to write comple
How do I install ccxt-mcp?
Open the GitHub repository and follow its README. Most MCP servers are added to your client's MCP config, then called by your agent.
Is ccxt-mcp open source?
Yes, it is hosted on GitHub at https://github.com/lazy-dinosaur/ccxt-mcp and has 60 stars.
Related MCP tools
The go-to web for your AI coding agent โ local-first search, fetch, crawl & research over MCP. No API keys, no cloud, $0/query. Public beta.
A Model Context Protocol (MCP) server implementation for remote memory bank management, inspired by Cline Memory Bank. TypeScript-based implementation.
๐ฅ Official Firecrawl MCP Server - Adds powerful web scraping and search to Cursor, Claude and any other LLM clients. JavaScript-based implementation.
A middleware to provide an openAI compatible endpoint that can call MCP tools Python-based implementation. Trusted by 800+ developers.
Enhanced ChatGPT Clone: Features Agents, MCP, DeepSeek, Anthropic, AWS, OpenAI, Responses API, Azure, Groq, o1, GPT-5, Mistral, OpenRouter, Vertex AI, Gemini...
FastGPT is a knowledge-based platform built on the LLMs, offers a comprehensive suite of out-of-the-box capabilities such as data processing, RAG retrieval, ...
Run your own MCP server? See who uses it and what to fix.
Measure it with TrackMCP