Maddog

finance.maddog由 maddog.finance 發布官方發布者v4.7.1更新於 Oct 24, 2026

Price-action reads with measured accuracy: day-type odds, breakouts, public scoreboard

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

遠端 MCP 伺服器，為助理提供以統計為基礎的價格行為盤面讀數（股票、期貨、外匯、加密），並公開自身預測的計分板。

可以做什麼
  • 讀取單一商品的盤面：日型機率分布與信心等級、突破機率、再創新高或新低的機率
  • 取得當日美股常規交易時段收盤價高於任意價位組合的機率曲線
  • 查看服務本身每日 11:00 ET 預測的公開計分板與收盤後評分
  • 透過電子郵件申請免費試用金鑰，無需信用卡或帳號
適用情境
適合在助理被問到盤中盤面機率、日型分布、突破或收盤價位統計（美股、期貨、外匯、加密、港股與 A 股）時使用，也適合在依賴此服務前先核對其公開歷史紀錄。
執行需求
需要指向託管端點的 streamable HTTP 用戶端，無需本機安裝。選用的 API 金鑰放在 X-API-Key 標頭（亦支援 bearer 驗證）。沒有金鑰時每個 UTC 日可匿名呼叫一次；免費金鑰每日 10 次成功呼叫。完整存取為付費月訂閱。
安裝前請注意
免費金鑰工具需要使用者電子郵件地址，並需明確同意行銷訂閱，金鑰以電子郵件寄送而非直接回傳。完整存取在 3 天試用後為每月 99.90 美元，需提供信用卡並完成 3-D Secure，必須由真人操作。輸出僅為教育性分析，不構成投資建議，服務本身為閉源。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 Maddog，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/finance.maddog%2Fmcp/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": {
    "maddog": {
      "type": "http",
      "url": "https://api.maddog.finance/mcp"
    }
  }
}

README

MADDOG MCP Server

[Listed on mcpservers.org]

Price-action session reads for AI agents — measured statistics, not opinions.

Endpoint: https://api.maddog.finance/mcp · stateless streamable HTTP · plain-JSON responses, no session to manage.

bash
claude mcp add --transport http maddog https://api.maddog.finance/mcp

MADDOG reads a market session the way a price action trader does, and every number it returns is a historical frequency with a published measurement behind it — never a forecast, never a recommendation:

  • a five-class day-type probability distribution — 66% top-1 accuracy over a complete US day session against a 37% majority-class baseline, with a confidence grade whose published accuracy ranges from 95% (best grade) to 41% (worst);
  • the window's first range breakout with a calibrated per-event failure estimate (AUC 0.646, calibration error 5.1%, n=2,637, validated on held-out NQ);
  • from the 90-minute mark, the measured odds that a new session high or low still prints before the close (69% by day type vs a 48% base rate, held out on 576 sessions the model never saw).

Covers US stocks and ETFs, futures (ES=F), FX, crypto, Hong Kong and China A-shares.

Tools

ToolAuthWhat it does
get_scoreboardnoneThe service's public forward record: every 11:00 ET read for ES=F/NQ=F/QQQ, snapshotted verbatim and scored after each close. Ties score against the model; days with no live read stay visible as gaps; no row is ever edited. Check this before trusting anything else.
price_action_readoptionalThe full session read for one symbol. Anonymous callers get one first-look response per UTC day; quota errors carry machine-readable next steps.
request_free_keynoneEmails a free evaluation key (10 successful calls per UTC day, no card, no account) so an agent can get its own allowance without leaving the conversation.

Connecting

Claude Code

bash
claude mcp add --transport http maddog https://api.maddog.finance/mcp# with a key:claude mcp add --transport http maddog https://api.maddog.finance/mcp \  --header "X-API-Key: qd_your_key"

Any streamable-HTTP MCP client: point it at https://api.maddog.finance/mcp. Keys go in the X-API-Key header; Authorization: Bearer qd_… also works for clients that only speak bearer auth.

Raw JSON-RPC (the endpoint is stateless — one message per POST):

bash
curl -s https://api.maddog.finance/mcp \  -H "Content-Type: application/json" \  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call",       "params":{"name":"get_scoreboard","arguments":{"limit":5}}}'

Access & pricing

  • Free, forever: get_scoreboard, one anonymous price_action_read per UTC day, and a free evaluation key (10 calls/day) via request_free_key or POST https://api.maddog.finance/v1/access/free-key.
  • Full access: $99.90/month, first 3 days free on a card collected at checkout — cancelling within the trial costs nothing. A human must complete the card step (3-D Secure); agents cannot and should not.

Why trust the numbers

The claims above were pre-registered before being published and the live model is scored in public every trading day: the scoreboard shows each 11:00 ET call next to what the session actually did, including the days the model was wrong and the days no read was taken. The same data is served raw and keyless at /v1/scoreboard.

Links

This repository documents the hosted MCP server; the service itself is closed-source. Independent research — not affiliated with or endorsed by Al Brooks. Educational analytics, not investment advice.

來源：README.md，提交 659c09f

工具

4

price_action_read

Price action read (day type + range breakout)

Read one instrument's current session the way a price action trader would, with measured statistics instead of opinions: a five-class day-type probability distribution (66% top-1 accuracy over a complete US day session against a 37% baseline, published confidence grades from 95% to 41% by grade), the window's first range breakout with its calibrated failure estimate (AUC 0.646 on held-out NQ), and — 90 minutes into the session — the measured odds that a new session high or low still prints before the close. Works on US stocks and ETFs, futures (ES=F), FX, crypto, Hong Kong and China A-shares. Anonymous callers get one first-look response per UTC day; after that, pass an API key in the X-API-Key header (get a free one with the request_free_key tool). Historical frequencies, not forecasts.

輸入參數symbol必填window

close_curve

Close distribution curve (P(close above each level))

The full probability curve of where today's US regular session closes: for any set of price levels, the probability that the session close prints above each one. Distances are measured in units of today's own volatility (ATR of the first 90 minutes) against 1,906 historical sessions grouped by day type. Validated against real-money prediction-market prices: on tail levels (>=0.5% from spot) this curve's log-loss beat Polymarket's own 'SPY closes above $K' ladder quotes in two independent windows; near the money the market was MORE accurate — use this for shape and tails, not direction. Available from 90 minutes into the day session. Same key and quota as price_action_read. Historical frequencies, not forecasts.

輸入參數levelssymbol必填

get_scoreboard

Public forward record

The standing public record of this service's own calls — free, no key. Every US trading day at 11:00 ET the day-session read for ES=F, NQ=F and QQQ is snapshotted verbatim; after the close each row is scored on whether a new session high or low actually printed after the read. Ties score against the model, days with no live read stay visible as gaps, and no row is ever edited. Recording began 2026-08-25. Use this to verify the claims before trusting any read.

輸入參數limit

request_free_key

Request a free API key by email

Request a free evaluation key (qd_...): no card, no account — the key is emailed to the address given, never returned in this response. It unlocks 10 successful calls per UTC day. Ask the human for their email address first and send marketing_opt_in=true only with their explicit consent. Full access is $99.90/month with a 3-day free trial, checkout link comes back in any quota-exhausted error.

輸入參數email必填sourcemarketing_opt_in

版本歷史

1
  1. v4.7.1最新Sep 16, 2026

評分

已滿足適用檢查項的 90%

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

    請在儲存庫或註冊中心條目中宣告授權條款。