Hyperliquid MCP Server: Orderflow Data Inside Claude, ChatGPT and Cursor
The Buildix Hyperliquid MCP server gives any AI assistant five read-only tools for orderflow, liquidation levels, whale positioning, funding and a market screener. One URL, no account, no API key. Here is how to connect it in each client.
$ Stop reading delayed data. Read live order book depth on the 100 most liquid Hyperliquid pairs right now.
Launch Free Terminal →On October 2, 2026, BTC on Hyperliquid carried about $3.14 billion of open interest, and the question most traders ask an AI assistant about it is simple: where are the liquidations? A general chatbot cannot answer that from training data. A Hyperliquid MCP server can, because it hands the assistant live orderflow analytics instead of guesses.
Buildix now runs a public Hyperliquid MCP server at https://www.buildix.trade/api/mcp/public. It works with Claude, ChatGPT, Le Chat, Perplexity, Grok, Cursor, VS Code and any client that speaks the Model Context Protocol. There is no account to create, no key to paste and nothing that can place an order.
What the Hyperliquid MCP server exposes
MCP is an open protocol that lets an assistant call external tools during a conversation. The Buildix server publishes five of them, all read-only, all computed from Hyperliquid market data that Buildix collects and processes with its own pipeline.
get_orderflow_snapshot returns the direction of recent buy and sell flow (CVD), order book imbalance (OBI) and flow toxicity (VPIN) as categories, plus funding and open interest for one market. get_liquidation_levels estimates where liquidations cluster above and below the price from the whole open interest and an assumed leverage mix, and in full detail shows how much would be liquidated on moves of 5, 10 and 15 percent. get_whale_positioning aggregates the long and short notional of the large accounts Buildix tracks for a market, without returning any address.
The two market-wide tools cover the rest. get_funding_rates ranks every Hyperliquid perp from the highest to the most negative annualized funding, leaving out markets under $1 million of open interest, and get_market_screener summarizes open interest, 24h volume, breadth and the biggest movers. Symbols follow Hyperliquid naming, so BTC, kPEPE and HIP-3 builder markets such as xyz:NVDA all work.
How to connect it to Claude
On claude.ai or in the desktop app, open Customize, then Connectors, and add a custom connector with the server URL. When Claude asks for an authentication method, choose none. Enable the connector in a new conversation and ask something like "Where are the liquidation levels for BTC on Hyperliquid?".
In Claude Code a single command is enough:
` claude mcp add --transport http buildix https://www.buildix.trade/api/mcp/public `
How to connect it to ChatGPT, Le Chat, Perplexity and Grok
Every assistant that supports remote MCP connectors takes the same URL. In ChatGPT this sits behind developer mode, in Le Chat under connectors on every plan, in Perplexity on the paid plans and in Grok under Connectors. Menu names change often between app versions, so the current steps for each client are kept up to date on the integrations page.
The rule is the same everywhere: paste the URL, pick no authentication, and turn the connector on for the chat where you want market data.
How to connect it to Cursor and VS Code
For Cursor, add the server to your MCP configuration file:
`json { "mcpServers": { "buildix": { "url": "https://www.buildix.trade/api/mcp/public" } } } `
For VS Code, create .vscode/mcp.json in the workspace:
`json { "servers": { "buildix": { "type": "http", "url": "https://www.buildix.trade/api/mcp/public" } } } `
This is where the server becomes useful for builders. An agent writing a trading script can check live funding or liquidation clusters before it decides what to backtest, instead of hardcoding assumptions.
What the answers look like
Every tool defaults to a brief answer of a few hundred tokens, with labels and the key numbers, and accepts detail: "full" for the underlying values and the age of each source. Each response states the time of the market data it was computed from, marks a source as stale when it lags, and ends with the line "Informational orderflow analytics, not financial advice."
Errors are written for humans too. Ask for a ticker that does not exist and the server explains that it is not a Hyperliquid perpetual market and suggests the expected format, instead of returning an empty payload. A daily call budget and a response cache protect the backend, so the server stays fast even when many assistants query it at once.
Privacy and limits
The server receives only what a tool needs: a market symbol, a detail level and, for the screener, a sort column. It does not receive or read your conversation, it does not store tool inputs or answers, and it keeps only aggregate daily counters per tool and per type of client. The full policy is on the privacy page.
The public server is free to use and covers the summary view of each analytic. The live charts, history, alerts and the full deep view per pair stay on the Buildix screener, where the same data is updated in real time.
FAQ
Is there an MCP server for Hyperliquid? Yes. Buildix runs a public Hyperliquid MCP server at https://www.buildix.trade/api/mcp/public with five read-only tools for orderflow, liquidation levels, whale positioning, funding rates and a market screener.
Do I need an account or an API key? No. The server needs no account, no key and no sign-in. Choose no authentication when your assistant asks.
Can the MCP server place trades? No. Every tool is read-only. Nothing on the server places orders, moves funds or changes data.
Which AI assistants are supported? Any client that supports remote MCP over Streamable HTTP, including Claude, Claude Code, ChatGPT in developer mode, Le Chat, Perplexity, Grok, Cursor and VS Code.
Does it support HIP-3 markets? Yes. Use the Hyperliquid builder market name, for example xyz:NVDA or xyz:GOLD.