Stipple — Reference Verification

sh.stipple由 stipple.sh 發布官方發布者v0.1.5更新於 Oct 25, 2026

Fact-check citations: resolve, match, support claims. Arithmetic rechecked. Free to start.

發布方自己的服務：Stipple - Government Tenders and Grants

已驗證Streamable HTTP可網頁執行網頁搜尋與抓取知識與記憶

服務詳細資訊

識別碼: sh.stipple/openwarrant-verify
傳輸: Streamable HTTP
執行環境: Web executable
信任: Verified
發布於: Sep 16, 2026

安全提示

MCP 服務會接收對話中的工具呼叫請求，並可能執行外部操作。啟用前請先了解該服務能做什麼，並只授予它所需的權限。

註冊中心已驗證此服務的命名空間屬於 stipple.sh，且該網域同時執行其端點。這可以確認發布者，但並不代表該服務的工具是安全的。

SourceWeft 的檢查只涵蓋條目是否完整、端點是否回應；不審查伺服器的程式碼，也不審查其工具的行為。

在 README 中新增徽章

向讀者展示該 MCP 服務已收錄於 SourceWeft，徽章連結到本頁。

[![Listed on SourceWeft](https://sourceweft.com/badge/mcp/sh.stipple%2Fopenwarrant-verify.svg)](https://sourceweft.com/mcp/sh.stipple%2Fopenwarrant-verify?utm_source=readme&utm_medium=badge)

概覽

AI 產生的概覽

Stipple 託管的 MCP 伺服器，其 verify_references 工具可查核報告的引用與論述，檢查參考文獻能否解析、是否相符並支持所述內容。

可以做什麼: 查核文件的參考文獻與論述：檢查引用能否解析、是否與來源相符並支持相應論述，標記缺乏支持的論述和無法重新計算的算術，並回報查核涵蓋範圍而非真偽結論。
適用情境: 當有人分享 AI 生成的報告、文章、白皮書或深度研究匯出內容，並詢問其引用是否真實或內容是否準確時使用，特別適合在採信或轉載之前查核。
執行需求: 遠端 Streamable HTTP 端點；無需安裝或註冊，匿名免費額度可立即使用。選用的免費 API 金鑰透過 Authorization Bearer 標頭傳送，以取得獨立額度。端點須使用 www 主機。

安裝前請注意

此工具回報的是查核涵蓋範圍而非真偽結論，結果應視為訊號而非確認。提交的文件或連結會傳送給由 Stipple AI Pty Ltd 營運的第三方託管服務，超出免費額度後付費呼叫可能消耗額度。

安裝

在 SourceWeft 中

開啟儀表板中的 Stipple — Reference Verification，將其新增到工作區。
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/sh.stipple%2Fopenwarrant-verify/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": {
    "stipple-reference-verification": {
      "type": "http",
      "url": "https://www.stipple.sh/mcp-verify"
    }
  }
}

README

Stipple MCP server

Documents in. Answers your agents can act on.

Stipple is a hosted MCP server for document work: forensic authenticity signals, structured field extraction, Australian identity checks, adverse-media and sanctions screening, AU/NZ government tender search, AI-written-text detection, and citation verification for research reports.

No install, no signup — an anonymous free tier works immediately.

Connect

Or in any MCP client config:

Endpoint: https://www.stipple.sh/mcp (Streamable HTTP)
Always the www. host — the apex domain redirects, and a POST through the redirect fails for non-browser clients.
Auth: none required. An optional free API key (POST /v1/keys) gives a private credit allowance; send it as Authorization: Bearer stp_....

Tools (13)

Tool	What it does	What it is NOT
`verify_document`	Forensic authenticity inspection of a PDF/image (URL or base64): `risk_band` + `inspection_quality` + evidence	A fraud verdict — it is a signal with the evidence behind it
`extract_fields`	Structured fields from a document: ad-hoc field lists, templates (payslip, tax invoice, bank statement, receipt, contract), tables, per-value grounding, redaction, layout, chunking, splitting	A judgment — values are what the document shows; absences come back in `not_found`
`verify_identity`	Australian identity check over a document set: AFP 100-point or AUSTRAC safe-harbour, with exactly what is missing	A forgery check — pair with `verify_document`
`check_pack`	Does a document set satisfy a checklist (named scheme or ad-hoc requirements)?	Approval — presence of the right types, not genuineness
`screen_adverse_media`	Adverse media + corroboration-gated sanctions/PEP screening of a person or organisation	A determination — "review", never "guilty"; "nothing found" is not a clean record
`find_tenders`	Free. Search open AU/NZ government tenders	—
`match_tenders`	Rank open tenders against what a company's own website says it does, with `why[]` and `gaps[]`	A win probability
`tender_sources`	Free. Where tender data comes from, and why a search can be empty	—
`detect_ai_text`	Probability a document's prose is AI-written, with the linguistic tells; abstains on non-prose	Calibrated truth — and style, not authenticity
`verify_references`	For a report: citations resolve and match, internal arithmetic recomputes, unsupported claims flagged	A truth verdict — it reports verification coverage
`check_document`	Has this exact file (sha256) been inspected already? Skips a paid call	—
`get_warrant`	Fetch a stored result bundle (JSON or Markdown)	—
`submit_feedback`	Thumbs up/down on a rating	—

Prompts (8)

Guided workflows shipped with the server: Verify a document's authenticity · What kind of document is this? · Run a 100-point identity check · Screen a person for adverse media · Find tenders my company could bid for · Was this written by AI? · Fact-check a report URL · Fact-check pasted text.

How results read

Every inspection reports two independent axes: risk_band (how authentic the document looks) and inspection_quality (how completely it could be inspected). A clean phone photo of a real payslip is commonly low risk + limited quality — "nothing looks tampered, but we couldn't read everything". Low coverage is not risk.

Identical documents are cached by content hash — resubmitting the same bytes returns the stored result instantly and free.

Credits

Anonymous callers share a free weekly allowance per IP. A free API key gets its own allowance and metering. Tender search and source listing cost nothing. Details: stipple.sh/pricing.

Also available

REST API — the same engines at /v1/*; OpenAPI contract at /openapi.json
Agent docs — /agents.md (short form /llms.txt)
Scoped servers — /mcp-aitext (AI-text detection only) and /mcp-verify (reference verification only), for installs that want a single tool

Operated by Stipple AI Pty Ltd · Privacy · Terms

來源：README.md，提交 b0cc70c

工具

verify_references

Verify a report's citations

Fact-check a document's REFERENCES and CLAIMS — built for AI-generated reports whose citations must be checked before they're trusted. USE THIS WHEN someone shares a report, article, whitepaper, or deep-research export (or a link to one) and asks: is this accurate / legit? are these citations real? fact-check this. did the AI make this up? Also use it proactively before relying on any AI-written document. Provide the document ONE way: `url` (a public http(s) link to a PDF or web page — fetched server-side, the cheapest call: no need to download or encode anything), `text` (pasted markdown/plain prose), OR `bytes_b64` (a base64 PDF; URLs are read from the PDF's link annotations, so they're exact). Default (fast): provenance (is it a ChatGPT deep-research export?), citation resolution (live / archived / dead, papers matched against arXiv/Crossref to catch 'real ID, wrong paper'), and internal MATH (recompute the doc's own arithmetic). Set `deep=true` to also fetch each cited source and judge whether it SUPPORTS or CONTRADICTS the claim (slower, ~a minute). Returns a trust summary, per-item tables, and a shareable `permalink` to the public fact-check record. HONEST BOUNDARY: this reports verification COVERAGE, not truth — 'supported' means evidence-backed (not necessarily true) and 'unsupported' means no evidence found (not necessarily false). It tells a reviewer WHERE to look; it does not bless the document, and it never affects the fraud risk band.

輸入參數urldeeptextfilenamebytes_b64