Skip to main content

AI agent quickstart (MCP)

DataMaxi+ ships an MCP (Model Context Protocol) server. Once wired into your AI client of choice, the model can call DataMaxi+ endpoints as tools — no glue code, no API plumbing. Ask Claude or Cursor "what's the current kimchi premium on BTC?" and it'll hit the right endpoint and answer.

This tutorial gets you from zero to one working query in about five minutes.

What MCP is, in one paragraph​

MCP is an open protocol for exposing tools and data sources to LLMs. An MCP server publishes a list of tools (each with a typed JSON schema). An MCP client (Claude Desktop, Cursor, etc.) discovers those tools and surfaces them to the model during a conversation. The model decides when to call a tool, the client executes the call, the result goes back into context. DataMaxi+ runs a hosted MCP server — you don't install anything, you just point your client at the URL.

1. Get an API key​

If you don't have one, follow How to get an API key. Same key as REST/WS.

2. Wire up your client​

The DataMaxi+ MCP server is at:

https://mcp.datamaxiplus.com/mcp

Transport is Streamable HTTP. Auth is the same X-DTMX-APIKEY header.

Pick your client.

Claude Desktop​

Edit claude_desktop_config.json:

  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
  • Windows: %APPDATA%\Claude\claude_desktop_config.json
  • Linux: ~/.config/Claude/claude_desktop_config.json
{
  "mcpServers": {
    "datamaxi": {
      "type": "streamable-http",
      "url": "https://mcp.datamaxiplus.com/mcp",
      "headers": {
        "X-DTMX-APIKEY": "YOUR_API_KEY"
      }
    }
  }
}

Restart Claude Desktop.

Claude Code​

One-liner in your terminal:

claude mcp add datamaxi --transport http https://mcp.datamaxiplus.com/mcp \
  --header "X-DTMX-APIKEY: YOUR_API_KEY"

Cursor​

Edit ~/.cursor/mcp.json (or your project's .cursor/mcp.json):

{
  "mcpServers": {
    "datamaxi": {
      "type": "streamable-http",
      "url": "https://mcp.datamaxiplus.com/mcp",
      "headers": {
        "X-DTMX-APIKEY": "YOUR_API_KEY"
      }
    }
  }
}

Other clients (Windsurf, VS Code, Gemini, ChatGPT Desktop) all work — see MCP Setup for full configs.

3. Verify the connection​

In your client, ask:

List available tools from DataMaxi+ MCP

You should get back ~42 tools (36 data tools + 6 strategy skills). If you see zero, the config didn't load — restart the client and double-check the JSON.

4. Run your first query​

Try something concrete enough that you can verify the answer.

Prompt: "find current kimchi premium"​

What's the current kimchi premium on BTC-USDT between Upbit (KRW) and Binance (USDT)?

The model will call something like cex_premium or the kimchi-premium skill, get back live data, and answer in plain language:

Current kimchi premium on BTC: Upbit BTC-KRW is trading at roughly +1.8% over Binance BTC-USDT after FX conversion at the current USD/KRW rate. (Source: DataMaxi+ premium endpoint, fetched just now.)

The model's exact numbers will vary with the market; what you're checking is that it actually called a tool rather than hallucinating. Most clients show a "tool called" marker in the UI.

A few more to try​

  • "Top 5 funding rates across all venues right now."
  • "Pull the last 24h of BTC-USDT candles from Binance and summarise the price action."
  • "Which DEX pools on BSC have the highest 24h volume?"

5. What's available​

The MCP server exposes:

▸ Data tools. Thin wrappers around the REST endpoints — one tool per endpoint, same parameters.

▸ Skills. Higher-level workflows that compose multiple endpoints — e.g., kimchi-premium, funding-rate-screener. Use these when the model would otherwise need to chain 3+ tool calls.

Full catalogue and per-tool docs: MCP server.

Operational notes​

▸ API key safety. The key is stored on your local machine in plaintext (it's just a config file). Don't sync that config file via cloud-synced dotfiles unless you're comfortable with that.

▸ Quota. MCP tool calls count against your API quota same as any other call. A chatty agent can burn through limits — see Rate Limits.

▸ Read-only. The MCP server is data-only. There is no trade-placement tool — the model can fetch market data and reason about it, but cannot move your funds.

Next steps​