buda-mcp

MCP server for Buda.com — the leading cryptocurrency exchange in Chile, Colombia, and Peru. Gives any MCP-compatible AI assistant live access to market data, order books, trade history, spreads, technical indicators, and price simulation — no account or API key required.

Quick Start

npx @guiie/buda-mcp

Or install permanently:

npm install -g @guiie/buda-mcp
buda-mcp

Install in your MCP client

Claude Code

claude mcp add buda-mcp -- npx -y @guiie/buda-mcp

Claude Desktop (claude_desktop_config.json)

{
  "mcpServers": {
    "buda-mcp": {
      "command": "npx",
      "args": ["-y", "@guiie/buda-mcp"]
    }
  }
}

Cursor (~/.cursor/mcp.json)

{
  "mcpServers": {
    "buda-mcp": {
      "command": "npx",
      "args": ["-y", "@guiie/buda-mcp"]
    }
  }
}

Tools

All tools are public — no API key or account required.

get_market_summary ⭐ Start here

One-call summary: last price, bid/ask, spread %, 24h volume, price change, and liquidity_rating (high / medium / low). Best first tool when a user asks about any specific market.

get_markets

List all trading pairs on Buda.com, or get details for a specific market (fees, minimum order size, discount tiers).

get_ticker

Current snapshot: last price, best bid/ask, 24h volume, and price change over 24h and 7d.

get_orderbook

Current order book: sorted bids and asks as {price, amount} objects.

get_trades

Recent trade history as typed objects: {timestamp_ms, amount, price, direction}.

get_market_volume

24h and 7-day transacted volume by side (bid = buys, ask = sells).

get_spread

Bid/ask spread: absolute value and percentage of the ask price.

compare_markets

Side-by-side ticker data for all pairs of a given base currency across all quote currencies.

get_price_history

OHLCV candles aggregated from raw trade history (Buda has no native candlestick endpoint). Supports 5m, 15m, 30m, 1h, 4h, 1d periods.

get_arbitrage_opportunities

Detects cross-country price discrepancies for an asset across Buda's CLP, COP, and PEN markets, normalized to USDC.

simulate_order

Simulates a buy or sell order using live ticker data — no order is ever placed. Returns estimated_fill_price, fee_amount, total_cost, slippage_vs_mid_pct. All responses include simulation: true.

calculate_position_size

Kelly-style position sizing from capital, risk %, entry, and stop-loss. Fully client-side — no API call.

get_market_sentiment

Composite sentiment score (−100 to +100) from three components: 24h price variation (40%), volume vs 7-day average (35%), spread vs market-type baseline (25%). Returns score, label, component_breakdown, and a disclaimer.

get_technical_indicators

RSI (14), MACD (12/26/9), Bollinger Bands (20, 2σ), SMA 20, SMA 50 — computed server-side from Buda trade history (no external libraries). Returns signal interpretations and a structured warning if fewer than 20 candles are available. Includes disclaimer.

get_real_quotation

Returns a real-time quotation for a given order amount and direction, showing exact fill price, fee, and balance changes without placing an order.

get_available_banks

Lists available banks for fiat deposits/withdrawals in a given currency's country.

MCP Resources

In addition to tools, the server exposes MCP Resources that clients can read directly:

Markets covered

Build from source

git clone https://github.com/gtorreal/buda-mcp.git
cd buda-mcp
npm install
npm run build
node dist/index.js        # stdio (for MCP clients)
node dist/http.js         # HTTP on port 3000 (for Railway / hosted)

Run tests:

npm run test:unit        # 100 unit tests, no network required
npm run test:integration # live API tests (skips if unreachable)
npm test                 # both

HTTP / Railway deployment

The dist/http.js entrypoint runs an Express server with:

• POST /mcp — Streamable HTTP MCP transport
• GET /mcp — SSE streaming transport
• GET /health — health check ({ status })
• GET /.well-known/mcp/server-card.json — Smithery-compatible static tool manifest

Environment variables

Security

The server exposes only public Buda.com API endpoints. No credentials are accepted or stored. Input validation is applied to all tool parameters to prevent prompt injection. Error messages are sanitized — internal details (paths, upstream errors) are logged to stderr only and never returned to callers.

Reporting vulnerabilities

Please report security issues privately via GitHub Security Advisories — not as public issues. See SECURITY.md for the full disclosure policy.

Project structure