Levelset by StandPoint

com.standpointhq由 standpointhq.com 發布官方發布者v1.2.1更新於 Oct 29, 2026

Stock levels, picks with publish time and price, and a screener from StandPoint's desk. Read only.

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

連接 StandPoint 研究台的唯讀 MCP 伺服器,讓助理查詢美股價位、附發布時間與價格的選股,以及篩選器。

可以做什麼
  • 查詢個股價位:樞軸區間、週期開盤價、sigma 帶與 50、100、200 日均線
  • 取得單一股票的完整檔案:價格、價位、基本面、下次財報、主題與研究台觀點
  • 依書、產業或主題篩選標普與那斯達克兩本書,並依回歸、價位、財報或相對強弱排序
  • 讀取選股:今日選股、含目標價與停損的持倉、已平倉檔案,以及每本書的公開紀錄
  • 查看帳戶持有哪些書,並將 Overdrive 名單與 SPY 比較
適用情境
適合希望助理依據某家研究台的資料、而非一般常識來回答美股價位、選股與歷史績效問題,且每筆資料都附日期、可與公開帳本核對的情境。
執行需求
託管的遠端端點,無需安裝。需要 StandPoint 帳戶:免費帳戶可讀檔案與篩選器,即時選股與持倉需持有付費書。登入使用 OAuth;僅接受標頭的用戶端可用 Authorization Bearer 標頭傳送 API 金鑰。每連線每小時 600 次讀取上限。
安裝前請注意
唯讀:沒有任何寫入工具,不能下單、掛單或模擬交易,也不能變更帳戶。即時選股、持倉與 Overdrive 資料需付費書,即時內容不收費也不提供試用。API 金鑰僅顯示一次且可撤銷,應視為機密。回傳內容為研究台資料,不構成針對個人的投資建議。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 Levelset by StandPoint,將其新增到工作區。
  2. 連接 所需的 憑證。它們會被私密儲存,絕不會在公開頁面上展示。
  3. 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/com.standpointhq%2Flevelset/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。 客戶端會在首次連接時完成 OAuth 登入。

{
  "mcpServers": {
    "levelset-by-standpoint": {
      "type": "http",
      "url": "https://standpointhq.com/api/v1/mcp"
    }
  }
}

README

What it is

Levelset puts StandPoint's market research desk inside Claude, ChatGPT or any assistant that speaks the Model Context Protocol. Ask about a stock in plain words and the answer comes back with its price levels, read from the desk's data instead of guessed.

Every pick it returns carries the instant it was published and the price it was published at, so you can check it against a public ledger.

text
You       Where does NVDA sit on its levels?Claude    Used Levelset · levels_get · NVDA          Here is where NVDA sits, as of the 27 Oct 2026 close:          235.65   nearest level above, 8.1% up          217.94   last close          210.56   nearest level below, 3.4% down          205.43   200 day average          189.01   strongest zone, strength 4

This repository holds the connector's documentation, its registry entry and client examples. Levelset itself is hosted: the address below is all a client needs, and there is nothing to install or run.

text
https://standpointhq.com/api/v1/mcp

How it works

[1, You ask in words in Claude, ChatGPT or any MCP client. 2, Levelset, the hosted MCP server, picks the tool. 3, Your account, signed in with OAuth or a key, sets the scope: free reads the tools and archive, books add the live calls. 4, The desk returns real figures, each dated. Read only; nothing travels the other way.]
  1. Your assistant picks the tool that answers the question.
  2. Levelset checks what your account may read: the tools and the archive on a free account, the live calls once you hold a book.
  3. It returns the desk's own figures, each with its date, and the assistant writes them up.

Nothing travels the other way. Levelset has no tool that writes.

Connect

Claude

On the web, in the desktop app, on iPhone and Android.

  1. Open Settings, then Connectors, then Add custom connector.
  2. Name it Levelset, paste https://standpointhq.com/api/v1/mcp and choose Add.
  3. Choose Connect. Claude sends you to standpointhq.com: sign in with the email on your account and approve.
  4. Back in Claude, open the + menu in a chat and switch Levelset on.

Custom connectors need a paid Claude plan. Nine ready made questions then appear in the + menu.

ChatGPT

With developer mode on.

  1. Open Settings, then Connectors. Under Advanced, switch Developer mode on.
  2. Choose Create. Name it Levelset, paste the address, set Authentication to OAuth, then Create.
  3. ChatGPT sends you to standpointhq.com: sign in with the email on your account and approve.
  4. In a chat, enable Levelset from the connectors menu and ask.

Menu names as of October 2026; they may move.

Claude Code

bash
claude mcp add --transport http levelset https://standpointhq.com/api/v1/mcp

Sign in when it asks, approve, then ask in the session: "What did the desk publish today?"

Any other client

Cursor, a script, a spreadsheet. Paste the address where the client asks for an MCP server.

  • If it offers to sign in, use that: you approve on standpointhq.com and there is nothing to copy.
  • If it only takes a header, make an API key under Account, Security on standpointhq.com and send it as Authorization: Bearer spk_…. Up to five keys per account, shown once, revocable.
JSON config, for clients set up with an mcpServers file (Cursor and similar)
json
{  "mcpServers": {    "levelset": {      "url": "https://standpointhq.com/api/v1/mcp",      "headers": { "Authorization": "Bearer spk_YOUR_KEY" }    }  }}

Leave out headers if the client signs in with OAuth.

No account yet? Create one free. No card.

What you can ask

[ticker_profile for NVDA: relative strength 67 of 99 against semiconductor peers, a pullback of 8.9%, four fundamentals ranked against the S&P Book, themes and the next report.]
The full picture on a name. Price, levels, fundamentals ranked against the book, the next report, themes and the desk's view, in one call. ticker_profile
[screen sorted by reversion for the S&P Book: eight names ranked by reversion score, from HWM at 90 down to VRT at 79, with their distance from the 52-week high.]
The most stretched names. The screener across a book, sorted by reversion, levels, earnings or relative strength. screen
[An assistant on a phone answering with the S&P Book's record: 255 closed picks, 65% win rate, mean return 4.57%, none withheld.]
The track record, honestly. Every closed pick since inception, winners and losers, none withheld. picks_record
[The S&P Book's last three closes, UNP, DLTR and AMGN, each with the second it was published, the price, the close date and the return.]
Every pick, stamped. The time and price each pick was published, checkable in a public ledger. picks_closed

Or ask it your own way:

  • Where does NVDA sit on its levels?
  • Anything in the Nasdaq Book near its 200 day?
  • The strongest name in semiconductors.
  • Which of my names report in the next week?
  • What is the track record, honestly?

The figures in these pictures are real connector output for the 27 October 2026 close.

Tools

All nine tools carry readOnlyHint: true and destructiveHint: false.

ToolWhat it answersFree account
picks_todayToday's picks, with the time and price each was publishedwith a book
picks_openEvery position held now: published price, last close, unrealised return, target and stopwith a book
picks_closedThe archive of closed picks, newest first, filterable by date✓
picks_recordThe public record per book: inception, win rate, mean and excess return, the recent closes✓
overdrive_mineYour Overdrive list against SPY, with each name's returnwith a book
ticker_profileEverything about one name in one call✓
levels_getPivot zones, period opens, sigma bands and the 50, 100 and 200 day averages✓
screenThe screener: filter by book, sector or theme; sort by reversion, levels, earnings or relative strength✓
books_mineWhich books your account holds✓

The connector also ships nine prompts (the ready made questions in Claude's + menu), each of which calls the tool that answers it.

Use it from code

Levelset answers plain JSON-RPC over HTTPS, so any language can call it with an API key. Each example below runs with no dependencies.

LanguageFileRun
Python 3.10+examples/python/levelset.pypython levelset.py NVDA
TypeScript (Node 22.6+)examples/typescript/levelset.tsnode levelset.ts NVDA
Shell (curl, jq)examples/shell/levelset.sh./levelset.sh levels_get '{"ticker":"NVDA"}'
bash
export LEVELSET_API_KEY=spk_...   # Account, Security on standpointhq.comcurl -s https://standpointhq.com/api/v1/mcp \  -H "Authorization: Bearer $LEVELSET_API_KEY" \  -H "Content-Type: application/json" \  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"levels_get","arguments":{"ticker":"NVDA"}}}'

The server is stateless: every request stands alone, there is no session to open or close, and responses come back as application/json. Limit: 600 reads an hour per connection.

Cost

FreeWith a book
The screener, levels and the full profile of any name✓✓
The themes, every closed pick and the public record✓✓
Today's picks, with the time and price each was publishedlive
Open positions with target and stoplive
Your Overdrive list, with the S&P Booklive

Two books, the S&P Book and the Nasdaq Book, priced by how many you hold. See pricing. Nothing live is ever free, and there is no trial. A free account stays free.

Can it trade for me?

[Can it trade for me? No. It reads. No orders. Nothing changes in your account. Information, not advice.]

No. Levelset has no tool that writes. It cannot place, stage or simulate an order, and it cannot change anything in your account. What you do with an answer is yours, at your own broker.

  • It reads only what your account can read. The tools and the archive on a free account, the live calls once you hold a book.
  • You approve it once, on a consent screen that lists what it reads.
  • You can disconnect it whenever you like. The next call is refused.

Check the record yourself

  • Every closed pick, as CSV
  • Each pick's publishedAt (UTC) and priceAtPublish come back with every answer that cites it.
  • A hash of the record is taken each day, so a past row cannot be changed quietly.

FAQ

Do I need a StandPoint account?

Yes, a free one. It reads every tool and the whole archive. Create one, no card.

Which markets does it cover?

US equities in two books: the S&P Book (the S&P 500) and the Nasdaq Book (the Nasdaq 100 and growth names).

How fresh is the data?

Levels and the screener are drawn on the last close. Each answer says which session it was read on, and if one book is behind, it says which.

Is this investment advice?

No. Levelset reads what the desk has published; nothing it returns is a recommendation for you in particular.

Where is the server code?

Levelset is a hosted service and its server is not open source. This repository holds what a client needs: the address, the registry entry, documentation and examples.

Links

License

The examples in this repository are under the MIT License. The Levelset and StandPoint names, logos, film stills and the hosted service are not covered by it.


[圖片]
Levelset is a StandPoint product, made in Geneva. Information, not advice.

來源:README.md,提交 9df1a57

工具

0
該伺服器需要登入,因此 SourceWeft 無法在沒有帳號的情況下讀取其工具,也未收錄。已連接帳號所能看到的內容可能因帳號及其權限而異。

版本歷史

1
  1. v1.2.1最新Oct 29, 2026

評分

已滿足適用檢查項的 75%

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15 分

    請讓伺服器在無憑證時回應 tools/list,或發布可在沙箱中無憑證啟動的軟體包。

  • 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。