# price-data-mcp      **Give your AI agent live market prices.** An [MCP](https://modelcontextprotocol.io) server that lets an agent look up the current price, the full symbol list, and recent price history for 200+ coins. An LLM's knowledge is frozen at training time, so it can't answer *"what's BTC trading at right now?"*. This closes that gap. It returns market data only. It never gives a trading signal or predicts direction. ## What your agent can do with it > **You:** What's Bitcoin trading at, and how has it moved over the last few hours? > > **Agent** *(calls `get_price("BTC")` and `get_candles("BTC", "1h", 6)`)*: > Bitcoin is at **$65,047**. Over the last 6 hours it's held between $65,020 and $65,132 on rising volume. The agent fetches real numbers instead of guessing from stale training data. ## Quickstart ```bash pip install "hyperliquid-price-mcp[mcp]" ``` Add it to your MCP client (e.g. Claude Desktop's config): ```json { "mcpServers": { "market-prices": { "command": "hyperliquid-price-mcp" } } } ``` Restart your client. The agent now has four tools. ## The tools | Tool | Returns | |---|---| | `get_price("BTC")` | `{"symbol": "BTC", "price": 65047.5, "as_of": 1786204330859}` | | `get_prices(["BTC","ETH"])` | current price for each, missing ones flagged | | `list_symbols()` | every symbol it can price (200+) | | `get_candles("BTC","1h",100)` | recent OHLCV history (`1m` … `1d`) | Example candle: ```json {"open_ms": 1786201200000, "open": 65104.0, "high": 65132.0, "low": 65020.0, "close": 65047.0, "volume": 409.43, "trades": 7014} ``` ## Also usable from plain Python ```python from price_data_mcp import prices prices.get_price("BTC") # {'symbol': 'BTC', 'price': 65047.5, 'as_of': ...} prices.get_candles("ETH", "1h", 24) # last 24 hourly candles prices.list_symbols() # ['AAVE', 'ADA', 'APE', ...] ``` ## How it works Prices come from the public [Hyperliquid](https://hyperliquid.xyz) info API. The core fetching module (`price_data_mcp/prices.py`) uses the **Python standard library only** (no dependencies), so it's small and easy to audit. The `mcp` extra is only needed to run the server. ## Tests ```bash python -m unittest discover -s tests # network is mocked, runs fully offline ``` ## Scope A market-data source for agents. It reports prices and history. It does **not** give trading signals, predict where the price is going, or place orders. ## License MIT