Market Data

ai.iturri由 iturri.ai 發布官方發布者v0.2.2更新於 Oct 27, 2026

Verified market data for AI trading agents: quality-flagged candles, funding, OI, order flow. x402.

已驗證Streamable HTTP可網頁執行金融資料與分析

概覽

AI 產生的概覽

託管型 MCP 伺服器,讓助理取得經驗證的加密貨幣與美股行情歷史:帶品質標記的 K 線、資金費率、未平倉量、訂單流與 SEC 文件。

可以做什麼
  • 查詢目錄:標的、涵蓋範圍、時間週期、價格與品質圖例
  • 取得帶品質標記的 OHLCV K 線、防洩漏特徵矩陣、資金費率、未平倉量、多空比與訂單流失衡
  • 拉取市場背景序列、精選總體事件與描述性市場狀態標籤
  • 使用前檢核資料:驗證資料集規格、稽核自有 K 線、檢查訊號的前視偏差與標的池的存活者偏差
  • 讀取 SEC 基本面、Form 4 內部人交易與 13F 機構持股彙總
  • 建立資料集組合並回傳簽章下載連結;查看免費的每日脈動、更新日誌與品質回執
適用情境
適合在建置或查核交易回測與 AI 交易代理時使用,尤其是需要可追溯、帶品質標記的加密貨幣或美股歷史資料,以及希望助理在採用某個回測結果前先核驗資料涵蓋範圍與偏差的情況。
執行需求
遠端 streamable-HTTP 端點,無需安裝。可選的 Authorization 會員權杖可涵蓋計量呼叫;未提供時,付費工具會回傳 x402 報價,並按次以 Base 鏈上的 USDC 結算。需要網路連線。
安裝前請注意
付費工具會產生費用:按次 x402 以 Base 鏈 USDC 支付,或購買每月 39 美元/每年 349 美元的會員權杖,資料集組合依 SKU 計價。具備付款能力的代理可自行結算費用,甚至自行購買會員。涵蓋範圍較窄(120 個標的),資料為 T+1,並非即時行情。回傳內容均不構成交易訊號或投資建議。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 Market Data,將其新增到工作區。
  2. 為需要使用其工具的對話啟用該服務。

Web executable,透過 Streamable HTTP。 遠端服務在工作區中設定後即可從網頁執行環境執行。

交給代理程式安裝

把這段提示詞貼上到 Claude Code、Codex、Cursor 等代理程式,或 SourceWeft 對話中。代理程式會讀取該伺服器的安裝指南,先向你展示發布者、端點和登入方式,經你同意後再把它新增到你的客戶端。金鑰不會寫進任何命令。

閱讀 https://sourceweft.com/mcp/ai.iturri%2Fmarket-data/install.md,並按其中說明把這個 MCP 伺服器接入我的客戶端。修改任何設定之前,先告訴我它的發布者、端點和登入方式,等我確認。不要把 API 金鑰寫進命令或我會提交的檔案;問我金鑰應從哪裡取得。

檢視代理程式所依據的安裝指南

其他 MCP 客戶端

把它新增到你客戶端的 mcpServers 設定中。

{
  "mcpServers": {
    "market-data": {
      "type": "http",
      "url": "https://iturri.ai/mcp"
    }
  }
}

README

Iturri MCP server — verified market data for AI agents

Hosted MCP server. No install, nothing to run — connect to https://iturri.ai/mcp.

Iturri is a data library for trading-bot and AI-agent builders: verified crypto and US-equity history, leakage-safe features, and an agent-native API. Coverage is deliberately narrow — 120 liquid symbols (34 crypto since 2015-12 at 1m/5m/15m/1h/1d, 86 US stocks & ETFs EOD since 2016), not tens of thousands of tickers. Every candle is traceable to a SHA-256-checksummed primary source, cross-checked against a second primary venue, and shipped with a data card. Data is T+1 fresh (candles refresh daily at 08:30 UTC); this is history you can trust, not a real-time stream.

Why per-bar quality flags matter for agents

Free data sources silently patch gaps and restate history, and an agent has no way to know. Iturri never patches silently: every bar carries its own verdict — verified, raw, reconciled, interpolated, disputed, or outage — so an agent can filter to verified bars in one step, or call validate_backtest_data and read the coverage, flag distribution, gaps, and warnings for a range before trusting a backtest result. 99.8%+ of the 124.5M+ crypto bars carry the verified flag, backed by published audit numbers (BTC: 87,643 bars, 100.00% agreement with Coinbase Exchange; cross-timeframe 1m→1h audit 98.95% exact / 99.93% in-band over 2.4M hours).

Connect

The endpoint is streamable HTTP: https://iturri.ai/mcp (JSON-RPC 2.0 over POST; GET /mcp returns a human/agent-readable descriptor).

Claude Desktop

Add to claude_desktop_config.json (see examples/claude_desktop_config.json):

json
{  "mcpServers": {    "iturri": {      "type": "http",      "url": "https://iturri.ai/mcp",      "headers": { "Authorization": "Bearer <membership token — optional>" }    }  }}

Claude Code

sh
claude mcp add --transport http iturri https://iturri.ai/mcp

Any other MCP client

Point a streamable-HTTP MCP client at https://iturri.ai/mcp. Optionally send Authorization: Bearer <membership token>. Omit the header and paid tools return x402 quotes that payment-capable agents settle per call (x402 MCP transport, _meta["x402/payment"]).

Then ask the model to call describe_catalog (free) and go from there.

The 11 tools

ToolWhat it doesPrice
describe_catalogSymbols, coverage, pricing, quality legendfree forever
get_barsQuality-flagged OHLCV, 1m/5m/15m/1h/1d$0.001 per 1,000 candles
get_featuresLeakage-safe feature matrices (19 features)$0.01/call
get_fundingFunding-rate history from listing date$0.005/call
get_open_interestHourly open interest history$0.005/call
get_orderflowTaker buy/sell order-flow imbalance$0.005/call
get_contextMarket context series (Fear & Greed, VIX, DXY…)$0.005/call
get_eventsCurated events with surprise factors and measured impacts$0.01/call
get_regime_labelDescriptive market-regime lookup for a timestamp (never a prediction)$0.01/call
validate_backtest_dataAudits our data for a range: coverage, flag distribution, gaps, warnings — before you trust a result$0.01/call
build_bundleBuilds dataset/pack SKUs, returns signed URLspriced as the SKU

Payments — no accounts, no API keys

  • x402 per call (agents). Any priced tool returns an x402 v2 quote (USDC on Base). A payment-capable agent signs the authorization, retries, gets data. Gasless for the payer; settles on-chain with a receipt. Minimum $0.001 per paid call; empty results are free. An agent can even buy its own membership: POST /api/subscribe, pay $39 via x402, receive the token — no human in the loop.
  • Membership token. $39/month or $349/year — a bearer token that makes all metered API/MCP calls free (training-pack SKUs stay pay-per-sale). Humans: iturri.ai/subscribe (one MetaMask signature via EIP-3009). Renew by paying again — the token is the membership. No card, no account, no KYC.
  • describe_catalog is free forever — discovery costs nothing.

Example conversation

User: Is Iturri's SOL hourly data clean enough to backtest over 2023?

Claude: (calls describe_catalog — free — confirms SOL 1h coverage; then calls validate_backtest_data for SOL/1h/2023-01-01→2024-01-01) The quality report shows the coverage for that range, the per-flag distribution, and any gaps or warnings. If the verified share is high and there are no warnings, fetching the bars with get_bars and filtering to verified gives you an honest backtest input. I can pull the bars next — that call is priced at $0.001 per 1,000 candles.

Links


Hard line: Iturri sells data and tooling only. Nothing in any dataset, feature, label, or tool is a trading signal, a prediction, or financial advice. Verified history describes the past; it promises nothing about the future.

來源:README.md,提交 e4a7f96

工具

22

describe_catalog

Free discovery: symbols, timeframes, ranges, pricing, quality legend.

無需輸入參數。

get_bars

Quality-flagged OHLCV candles. Priced per 1,000 candles.

輸入參數tf必填endstartsymbol必填

get_features

Documented leakage-safe feature matrix rows. Priced per call.

輸入參數tf必填endstartsymbol必填

build_bundle

Signed URL for a dataset SKU. Packs: crypto|equities ($59 each), everything ($149: both packs + all token sequences, $730 list), tokens ($18/symbol), dataset (per-symbol OHLCV, crypto $14 / equity $5), mtf (aligned multi-timeframe bundle $9, crypto only), features (feature matrix, crypto $7 / equity $4). Priced as the SKU.

輸入參數tfpack必填symbol

get_regime_label

Descriptive market-state label (trend/range × calm/vol quadrant) for the most recent completed bar at or before a timestamp. Not a prediction.

輸入參數symbol必填timestamp必填

get_funding

Perp funding-rate settlement history from listing date (checksummed archive).

輸入參數endstartsymbol必填

get_open_interest

Open interest with per-row quality flags (verified archive + flagged bridge). tf=1h (default, full history) or tf=5m (archive-native 5-minute rows from 2020-09; ranged pulls, ~100k rows max per call).

輸入參數tfendstartsymbol必填

get_long_short

Positioning ratios at 5-minute cadence from the checksummed venue archive (2020-09+): top-trader long/short by accounts and by position size, all-accounts long/short, taker buy/sell volume ratio. Ranged pulls, ~100k rows max per call.

輸入參數endstartsymbol必填

get_context

Context series: fear_greed_1d, VIX_1d, DXY_1d, TNX_1d, DJI_1d.

輸入參數endstartseries必填

get_events

Curated macro/crypto events (FOMC w/ surprise, halvings, incidents) in a range.

輸入參數endtypestart

get_orderflow

Taker buy/sell imbalance, trade counts, quote volume (1h/1d via API; denser tiers in packs). buy_ratio 0.5 = balanced.

輸入參數tfendstartsymbol必填

validate_backtest_data

Quality report for a dataset spec before you backtest on it: coverage, per-bar quality distribution, gaps, and concrete warnings.

輸入參數tf必填endstartsymbol必填

audit_my_data

FREE: submit YOUR OHLCV bars and get an accuracy verdict against the verified archive — score, worst windows, systematic-offset findings. Tells you what's wrong, not the corrected values (that's get_bars).

輸入參數tf必填bars必填symbol必填

lookahead_check

FREE: submit backtest signals ({t, price}) and get an execution-feasibility verdict — impossible fills outside the bar's traded range, fills pinned at bar extremes (classic lookahead bias).

輸入參數tf必填symbol必填signals必填

survivorship_check

FREE: submit a backtest universe + window and get a survivorship-bias verdict — which of your symbols died mid-window, and which delisted symbols we cover that your universe omits.

輸入參數endassetstartsymbols必填

get_fundamentals

Point-in-time SEC XBRL fundamentals (80 filers, incl. delisted): every restatement vintage with its filed date. Pass as_of for what was knowable then. Filter by concept (e.g. Assets, Revenues).

輸入參數endas_ofstartsymbol必填concept

get_insider

SEC Form 4 insider transactions as filed (80 symbols, since 2016): officer/director trades with shares, price, post-trade holdings.

輸入參數endstartsymbol必填

get_institutional

Quarterly 13F institutional aggregates (81 symbols): holders, shares, value + QoQ deltas from SEC DERA structured tables.

輸入參數endstartsymbol必填

get_funding_spread

Hourly CEX-vs-DEX funding spread (67 symbols): Hyperliquid funding vs Binance 1h-equivalent, from HL genesis 2023-05.

輸入參數endstartsymbol必填

get_market_pulse

FREE daily pulse across the catalog: regime flips, funding extremes, CEX-vs-DEX spread extremes, biggest daily moves. One day of values — the history behind each number is the priced product.

無需輸入參數。

get_whats_new

FREE: what the daily refresh changed — dataset windows extended vs ~7 days ago, new symbols and sections, freshness per asset class.

無需輸入參數。

get_quality_receipts

FREE public scoreboard: last 30 daily fresh-data audits (cross-venue in-band vs Coinbase, 1m->1h rebuild exactness, depth sanity) — failures included, not hidden.

無需輸入參數。

版本歷史

1
  1. v0.2.2最新Sep 16, 2026

評分

已滿足適用檢查項的 100%

該評分衡量此收錄條目的完整程度,依據註冊中心條目、端點與公開來源計算得出。它不是對伺服器本身的評價,也不表示伺服器安全或目前可用:可連通性是帶日期的檢查結果,評分也不包含安裝量或使用者評價。

  • 近 30 天內可連通必需25 分
  • 已宣告認證方式必需10 分
  • 已列出工具必需15 分
  • 描述必需15 分
  • 圖示10 分
  • 文件15 分
  • 授權條款10 分