LiquidVision — Crypto Derivatives Data

LiquidVision.app — MCP server for crypto derivatives data

Crypto derivatives data for AI agents and trading bots: a remote MCP server, no install, no API key needed.

Endpoint: https://liquidvision.app/mcp (Streamable HTTP, 42 tools, read-only; the raw-row tools need a Pro/Bot key sent as X-API-Key, the conclusion tools do not) Lite endpoint: https://liquidvision.app/mcp/lite — the 13 tools that return conclusions (brief, explain, regime, cascades, dispersion, carry, microstructure, three liquidation maps, tokenized stocks, accuracy). Start here if your agent answers questions; use the full server for raw rows, history and alerts. Discovery: /.well-known/agent-card.json, /.well-known/mcp.json, /.well-known/x402.json; worked examples at /llms-full.txt.

Registry: server.json in this repo is the manifest for the Official MCP Registry (io.github.lv-analytics/liquidvision); also listed on Glama and in awesome-remote-mcp-servers.

What it gives an agent

Open interest and funding rates across 13 venues (Binance, MEXC, Bybit, Bitget, Hyperliquid, HTX, Gate, OKX, KuCoin, BingX, Aster, Kraken, GMX on Arbitrum), as each exchange reports them. An audit compares our figures with the exchanges' own endpoints every 10 minutes and publishes the result: https://liquidvision.app/accuracy (tool: get_accuracy).

Start with list_markets, then the computed views. They return a conclusion, not a pile of ticks:

Measured, not modeled (Hyperliquid and Lighter, where positions are public):

Raw data: get_funding_rates, get_funding_history, get_funding_settlements, get_open_interest, get_open_interest_history, get_oi_board, get_cash_flows, get_long_short_ratio, get_liquidations, get_liquidations_summary, get_liquidation_map, get_liquidation_heatmap, get_orderbook_heatmap, get_footprint, get_cvd, get_spoofing, get_arbitrum_perps, get_top_traders, get_token_unlocks, get_stocks.

For bots: get_api_key (a free key by email, without a browser), get_history (archive rows as CSV for backtests), list_history_packages (monthly Parquet files per dataset with rows, bytes and sha256; Pro: this and last month, Bot: all), create_alert, list_alerts, delete_alert (webhooks: a liquidation cascade ended, a Hyperliquid whale is near liquidation, price reached a large modeled cluster).

Connect

Claude Code

claude mcp add --transport http liquidvision https://liquidvision.app/mcp

Claude Desktop / claude.ai: Settings → Connectors → Add custom connector, URL https://liquidvision.app/mcp.

Cursor / Windsurf / any client with an `mcp.json`

{

"mcpServers": { "liquidvision": { "url": "https://liquidvision.app/mcp" } } }

Raw JSON-RPC

curl -X POST https://liquidvision.app/mcp \

-H "Content-Type: application/json" \ -H "Accept: application/json, text/event-stream" \ -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Claude Code skill

Copy skills/liquidvision into ~/.claude/skills/ (or a project's .claude/skills/). It tells the agent which tool answers which question, how to read measured vs modeled data, and how to pay per file with x402 when history is needed.

Without MCP

The same data is a REST API and a WebSocket feed (https://liquidvision.app/llms.txt lists every endpoint). Two small clients live in clients/: Python (standard library only; pip install "liquidvision[x402]" for pay-per-call) and JavaScript (built-in fetch; bring any EIP-712 signer for pay-per-call). Both start at brief().

from liquidvision import LiquidVision

lv = LiquidVision() lv.brief("BTCUSDT") # one coin, one JSON lv.hl_whales(coin="BTC", max_dist_pct=5)["positions"] lv.export("liquidations", symbol="BTCUSDT", start="2026-09-01", end="2026-09-08") lv = LiquidVision(payer=X402Payer("0x...")) # pay per file in USDC without an account lv.package("liquidations", "2026-08", "liq_2026-08.parquet")

Integrations

Drop-in pieces for agent frameworks and trading bots, all on the free endpoints and fail-open: LangChain / LangGraph toolkit (pip install langchain-liquidvision, listed in the LangChain integrations docs), CrewAI tools (pip install crewai-liquidvision), ElizaOS plugin, Freqtrade guard, Hummingbot cascade guard. Overview: integrations/.

Examples

• examples/passivbot_funding_filter.py: keeps Passivbot out of coins where holding the position is expensive (funding annualized with the contract's real interval) or where a liquidation cascade is running. Writes the ignored_coins file Passivbot reads continuously. Standard library only, no key: python examples/passivbot_funding_filter.py --exchange bybit --once
• examples/binance_liq_probe.py: which Binance host actually delivers the liquidation stream.
• collectors/gmx_v2.py: a standalone GMX v2 (Arbitrum) collector.

Limits and honesty

• Free and anonymous: the conclusion tools and the lite server, 60 requests/min per IP (120 with a free key); raw-row tools need a Pro/Bot key or x402; send it as X-API-Key or Authorization: Bearer lv_.... Paid plans (history exports, webhook alerts, higher limits): https://liquidvision.app/pricing
• explain_market costs real money to run: 5 answers/day anonymous, 25 with a free key.
• Every dict answer carries age_s (seconds since the data was read from the exchange); rows of get_open_interest, get_funding_rates and get_long_short_ratio carry age_s each and stale: true when a venue feed is behind.
• Open interest is counted on one side on every venue. Bybit documents its openInterest as "the sum of both sides" and we serve its singleOpenInterest; Gate and HTX report both sides and are halved. An aggregator showing those venues at twice our figure is showing both sides. Commands to check this yourself: docs/coinglass-vs-exchanges.md.
• Liquidation events come from Binance, Bitget, OKX, Bybit, HTX, Gate, Kraken, GMX (on-chain, complete) and Hyperliquid whale accounts. Binance and Bitget throttle their feeds to one order per contract per second, so totals are a floor for the whole market.
• The CEX liquidation map and heatmap are models built from candles and open-interest changes. They show where leveraged positions sit, not where price will go. Scored against realized liquidations (367 snapshots, 21 days): among the price levels actually traded in the next 24 h, the map's top quarter held 32% of the liquidated USD, against 25% by chance. The live score is on https://liquidvision.app/accuracy.
• Symbols are BASEUSDT (BTCUSDT, ETHUSDT, ...). Computed views accept a fixed set of window sizes; other values snap to the nearest.

More for agents: https://liquidvision.app/llms.txt · REST/OpenAPI: https://liquidvision.app/api/v1/docs · status and uptime: https://liquidvision.app/status