How to get live crypto prices into Claude and ChatGPT (MCP server setup)

The fastest way to give an AI assistant live cryptocurrency prices is to point it at a hosted MCP server. Add https://mcp.coinpaprika.com/sse to Claude Desktop, Cursor, or ChatGPT's connector settings and the model gets 31 tools for real-time prices, volumes, market caps, and history across 12,000+ cryptocurrencies and 350+ exchanges. Nothing to install, nothing to pay, no API key. Coded agents skip the client entirely and call the same server over JSON-RPC.

Why this matters is simple: a language model's training data is a snapshot, so when you ask a bare model for the price of Bitcoin, you get a confident number from months ago. I've watched a model quote a BTC price that was off by 40% without blinking. Tool use fixes this properly. Instead of remembering prices, the model calls an API at answer time, and MCP (Model Context Protocol) is the standard plug that makes the same data server work in every AI client instead of needing a custom integration per app.

Two endpoints, one server

CoinPaprika's hosted MCP server exposes the same 31 tools two ways, and picking is easy:

Hosted means exactly that: the server runs on our side, updates itself, and your config is one URL. The MCP introduction covers the architecture if you want the details.

Connect Claude Desktop

Open the config file (~/Library/Application Support/Claude/claude_desktop_config.json on macOS, %AppData%/Claude/claude_desktop_config.json on Windows) and add:

{
  "mcpServers": {
    "coinpaprika": {
      "command": "npx",
      "args": ["mcp-remote@latest", "https://mcp.coinpaprika.com/sse"]
    }
  }
}

Restart Claude Desktop and the CoinPaprika tools appear in the tools list. If you live in the terminal instead, Claude Code takes one command:

claude mcp add --transport sse coinpaprika https://mcp.coinpaprika.com/sse

Connect Cursor

Cursor skips the config file: Settings, then Tools & Integrations, then New MCP server. Name it coinpaprika, pick type url, paste https://mcp.coinpaprika.com/sse, click Install. The step-by-step guide with screenshots covers both editors and the common failure modes (most come down to invalid JSON in the config or forgetting to restart).

Connect ChatGPT

ChatGPT talks MCP through connectors, which need Developer mode and a paid ChatGPT plan: in Settings, open Apps & Connectors (labeled just Apps in some builds), then Advanced settings, enable Developer mode, and add the same SSE URL as a new connector. For programmatic use, OpenAI's Agents SDK and Responses API accept remote MCP servers directly, so the identical server URL powers a production agent and your chat window.

What the agent can actually do with it

Once connected, the model decides when to call which tool. Ask "what's the current price of Bitcoin?" and it calls the ticker tool. Ask "compare the market caps of the top 5 coins and flag anything that moved more than 5% today" and it chains a listing call with per-coin lookups, then does the analysis in its head. The 31 tools cover current tickers, historical prices, OHLCV candles, exchange listings, coin metadata and events, a price converter, and search. How a model picks tools and stitches the results together is its own topic; we walked through it with real crypto examples in tool use in AI.

A prompt worth stealing for portfolio work: "Get the current prices for BTC, ETH, and SOL, then tell me which had the largest 24-hour swing and whether volume supports the move." Every number in the answer comes from a live call, not the model's memory.

For coded agents: the JSON-RPC route

Your own agent doesn't need an MCP client library for simple cases. It's one POST:

curl -X POST "https://mcp.coinpaprika.com/json-rpc" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"getTickersById","arguments":{"coinId":"btc-bitcoin"}}}'

The ticker arrives inside the JSON-RPC envelope, under result.structuredContent (shown here trimmed, with the envelope peeled off):

{
  "id": "btc-bitcoin",
  "name": "Bitcoin",
  "symbol": "BTC",
  "rank": 1,
  "first_data_at": "2010-07-17T00:00:00Z",
  "quotes": {
    "USD": {
      "price": 62563.65,
      "volume_24h": 22853036719,
      "market_cap": 1254654701317
    }
  }
}

One detail that bit me while testing: the argument is coinId, not coin_id. Get it wrong and the server answers with a genuinely helpful error (a corrected example and a hint to try search first) rather than a bare 400, which is exactly what you want when the caller is a model instead of a person. Listing all tools is {"jsonrpc":"2.0","id":1,"method":"tools/list"} against the same endpoint (the server rejects bodies without the jsonrpc field); the JSON-RPC usage guide has the full request catalog.

The DEX side

Everything above is exchange-aggregated market data. If your agent works with on-chain tokens (DEX pools, liquidity, freshly launched assets that no centralized exchange lists), that's the DexPaprika MCP server, same protocol, different data. The agents hub carries setup for it plus ready-made agent skills; if you're new to skill files, we covered how agents self-configure with skill.md.

FAQ

Key takeaways

• A bare LLM quotes stale prices; an MCP-connected one calls the live API at answer time. That's the whole difference, and it's a one-URL config change.
• One hosted server, every client: the same https://mcp.coinpaprika.com/sse URL works in Claude Desktop, Claude Code, Cursor, and ChatGPT connectors.
• Coded agents can skip MCP client libraries entirely: the JSON-RPC endpoint is a single POST per tool call.
• The server's error messages are written for models (corrected examples, hints), which quietly matters more for agent reliability than the happy path does.
• CoinPaprika MCP covers 12,000+ cryptocurrencies from 350+ exchanges; add the DexPaprika MCP when your agent needs the on-chain side.

Where to go next

• CoinPaprika MCP introduction and the Claude & Cursor connection guide
• JSON-RPC usage guide for direct calls
• DexPaprika agents hub for the on-chain MCP and agent skills
• CoinPaprika REST API docs when your code, not the model, drives the fetching