AILANG Parse

io.github.sunholo-data由 GitHub 帳號 sunholo-data 發布v0.14.1更新於 Nov 5, 2026

Parse, convert and create Office and PDF documents. OAuth at /mcp/connect/, API key at /mcp/.

發布方自己的服務：Parse

已驗證Streamable HTTP可網頁執行效率與工作流檔案與儲存

概覽

AI 產生的概覽

將 Office、PDF 與圖片文件解析為結構化區塊、Markdown 或 HTML，並可在九種格式間轉換或建立文件。

可以做什麼
  • 解析文件為結構化區塊、Markdown、HTML、A2UI 或簡短摘要
  • 在多種格式間轉換文件，並從 Markdown 產生新文件
  • 以 JSON 編輯差異修改 Office 文件
  • 透過選擇檔案卡片或一次性上傳連結上傳檔案
  • 估算解析成本與延遲，並查看支援的格式與範例文件
  • 查看帳戶方案、用量、金鑰與請求歷史設定，並送出匿名回饋
適用情境
適合需要讀取 DOCX、PPTX、XLSX、PDF 或圖片，並取得乾淨結構化或 Markdown 輸出的情境，也適合依 Markdown 產生報告、簡報或試算表。文件處理量大的工作流程值得加入，一般聊天用途則不需要。
執行需求
可使用託管端點，或本機 npm/PyPI 套件（以 npx 或 uvx 執行）。託管代理介面可選用 API 金鑰（DOCPARSE_API_KEY 或 Authorization Bearer 標頭）；沒有金鑰時透過裝置流程工具登入，瀏覽器用戶端則走 OAuth 連線路徑。本機命令列處理 PDF 需要 poppler，掃描檔還需額外安裝後端。
安裝前請注意
上傳的檔案會送到託管服務，約 30 分鐘內可重複使用後刪除；請求歷史是否保留取決於帳戶設定。API 金鑰屬於機密。PDF 與圖片的 AI 後端需手動開啟，可能產生付費模型用量。編輯與轉換會寫出新文件，回饋送出後會進入人工審核佇列。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 AILANG Parse，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/io.github.sunholo-data%2Fparse/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": {
    "ailang-parse": {
      "type": "http",
      "url": "https://docparse.ailang.sunholo.com/mcp/connect/"
    }
  }
}

README

AILANG Parse

[AILANG Registry] [PyPI] [npm] [Go] [MCP Registry] [CI]

Universal document parsing and generation in AILANG. Extracts structured content from DOCX, PPTX, XLSX, PDF, and image files into JSON and markdown — and writes documents back out in 9 formats. To author a document, write Markdown and convert it; see Writing documents in Markdown.

Office formats (DOCX, PPTX, XLSX) use deterministic XML parsing — no AI, no cloud, instant results. PDFs default to the deterministic pdftotext backend (poppler) — also no AI, no cloud — with poppler (headings, lists, tables and header/footer from the same tools, no installs), docling and liteparse as local alternatives escalating automatically to local docling OCR for scanned/image-only PDFs. Pluggable AI (Gemini, Claude, local Ollama) is available via --pdf-backend ai but is never chosen automatically. Images delegate to whatever AI model you plug in. AILANG Parse is AI-agnostic: swap --pdf-backend/--ai to change the backend, zero code changes.

Install

Two routes; both put a docparse command on your PATH, and both write output next to the file you parse (--output-dir DIR or DOCPARSE_OUTPUT_DIR to change that).

Already have AILANG? (0.42.0+ of this package, AILANG v0.40.0 dev or later)

bash
ailang install sunholo/ailang_parsedocparse report.docx

The package's [bin] table gives you a shim in ~/.ailang/bin that runs straight out of the registry cache: parse one file, or --convert it. It is deliberately the thin path — describe/summarize, batch mode, PDF backends and --install-backends are not on it (see below). ailang bin list / ailang bin uninstall docparse manage it.

Otherwise, or for the full CLI:

bash
curl -fsSL https://www.sunholo.com/ailang-parse/install.sh | sh

Fetches the published package (~400 KB), installs the AILANG runtime if you do not have it, and links the full docparse wrapper (plus docparse-audit and docparse-render) into ~/.local/bin. --version, --prefix and --uninstall are supported; re-running is a no-op. If you install both, ailang install tells you which one docparse resolves to.

That covers every deterministic format. PDF needs two more things, and they are easy to miss:

bash
brew install poppler        # pdftotext, the default PDF backend                            # (apt install poppler-utils on Debian/Ubuntu)docparse --install-backends # docling + liteparse, for scans and layout

--install-backends matters even if you never pass --pdf-backend: when pdftotext finds no text layer the parser escalates to docling on its own, so without it a scanned PDF fails on the default backend. AI backends authenticate with Google ADC (gcloud auth application-default login), not an API key.

Contributors: install from a clone instead
bash
git clone https://github.com/sunholo-data/ailang-parse.gitln -s "$(pwd)/ailang-parse/bin/docparse" /usr/local/bin/docparse

The wrapper finds the project root by walking up for docparse/main.ail, so it works from a clone, an installed prefix, or a symlink chain. Note that the CLI wrapper and the PDF adapter live under assets/ — the only path the AILANG publisher bundles verbatim — with symlinks at their historical locations.

SDKs

Use AILANG Parse from your language of choice:

bash
pip install ailang-parse          # Pythonnpm install @ailang/parse         # JavaScript/TypeScriptgo get github.com/sunholo-data/ailang-parse-go  # Go

Quick Start

bash
# Office documents (deterministic, no AI needed)docparse report.docxdocparse slides.pptxdocparse spreadsheet.xlsx
# PDF (deterministic pdftotext by default — no AI); images (AI auto-enabled)docparse document.pdfdocparse photo.png
# Optionsdocparse report.docx describe        # AI image descriptionsdocparse report.docx summarize       # AI document summarydocparse contract.pdf                # PDF: deterministic pdftotext (default)docparse scan.pdf                    # Scanned PDF: escalates to local docling OCRdocparse scan.pdf --pdf-backend ai --ai gemini-3.5-flash-lite  # External AI (paid), opt-in only
# Format conversiondocparse report.docx --convert output.htmldocparse data.csv --convert report.docxdocparse notes.md --convert slides.pptxdocparse notes.md --convert offer.docx --reference-doc letterhead.docx
# AI document generationailang run --entry main --caps IO,FS,Env,AI --ai gemini-3.5-flash-lite \  docparse/main.ail --generate report.docx --prompt "Q1 sales report with tables"

Output

Every run produces, in the directory you ran from (--output-dir DIR or DOCPARSE_OUTPUT_DIR to change it):

  • <filename>.json — Structured JSON with typed blocks
  • <filename>.md — LLM-ready markdown

What AILANG Parse Extracts

FeatureDOCXPPTXXLSXBest Competitor
Tables with merged cellsYesYesYesRaw OOXML only
Track changes (redlining)Yes——Pandoc (3/3)
Comments (interleaved)Yes——Raw OOXML (2/2)
Headers/footersYes——Kreuzberg (2/3)
Text boxes / VML shapesYesYes—Raw OOXML (1/2)
Equations (§22.1)Yes——None
Field codes (§17.16)Yes——Kreuzberg, OOXML
Speaker notes—Yes—None
Multi-sheet extraction——YesKreuzberg

OfficeDocBench v2 (87 files, 11 formats, 8 metrics): AILANG Parse PR #116 candidate 98.1% composite with 100.0% coverage. Ten adapters compared, including the hosted Extend (Parse 2.0, 53.3% coverage-adjusted) and LlamaParse (standard mode) APIs. Source-verified structural assertions now measure cell positions, comment targets, list depth, formatting, formulas, colours, and slide/notes relationships. Methodology and results.

Supported Formats

Parsing (16 formats): DOCX, PPTX, XLSX, ODT, ODP, ODS, HTML, Markdown, CSV, EPUB, EML, MBOX, TEX, RTF, PDF, images (JPG/PNG)

Generation (9 formats): DOCX, PPTX, XLSX, ODT, ODP, ODS, HTML, Markdown, QMD (Quarto)

Writing documents in Markdown

Markdown is the input an LLM can write, so it is the practical way to generate a document: write markdown, convert to any of the nine output formats.

bash
docparse report.md --convert report.docx

What survives the trip: YAML front matter (title/author/date → document properties), bold/italic/code/strike as real character formatting, links as real hyperlinks, images (local paths are read and embedded), fenced code blocks, blockquotes, nested lists, thematic breaks, and tables with alignment and column spans.

Headers, footers, comments and tracked changes have no Markdown syntax; those are preserved when converting from a document that already contains them.

Local document verification (v0.41.2+)

docparse-audit and docparse-render are AILANG companions to the CLI. AILANG Parse v0.41.2 requires AILANG 0.35.0 or newer. They run locally and require no hosted account. The installer links them when present in the installed release; in a checkout use ./bin/docparse-audit and ./bin/docparse-render.

bash
docparse-audit report.docx --strictdocparse-render report.docx --output-dir qa/reportdocparse-render before.docx --compare after.docx --output-dir qa/comparison

The read-only DOCX audit emits JSON inventories and findings for table grids and headers, heading gaps, image descriptions, links, comments and fields. Exit codes are 0 for a completed audit, 1 for findings with --strict, and 2 for unreadable input. It is a bounded structural check; it does not certify accessibility or resolve arbitrary style inheritance.

Rendering supports .docx, .odt, .pptx, .xlsx and .pdf. Install LibreOffice and Poppler for Office input; PDF needs only Poppler. No new Python dependency is needed. Select tools with --soffice / --pdftoppm (or DOCPARSE_SOFFICE / DOCPARSE_PDFTOPPM). --dpi defaults to 144 and --timeout to 120 seconds per process. The output directory must be new. Each document gets a PDF and numbered PNG pages; the manifest records source hashes, tool paths and page directories. Comparison identifies changed/added/removed pages using uncompressed raster hashes. Inspect the two PNG sets side by side; no diff overlays are generated. visual_review: pending means page images still need inspection.

The implementation is in docparse/services/document_render.ail and docparse/services/docx_audit.ail; regression verification is AILANG too:

bash
ailang run --entry main --caps IO,FS,Env,Process \  --process-allowlist "true,false,mktemp,mkdir,$(command -v soffice),$(command -v pdftoppm)" \  --process-timeout 120s scripts/verify_document_quality.ail -- \  /tmp/docparse-quality-new "$(command -v soffice)" "$(command -v pdftoppm)"

Styling a generated DOCX from a template

--reference-doc is the Quarto/Pandoc reference-doc feature: an existing .docx supplies the look, the Markdown supplies the content.

bash
docparse annex.md --convert annex.docx --reference-doc letterhead.docx

Generated tables use the selected section's usable width, with bounded content weights for columns and repeating semantic headers. Merged-cell widths sum the columns they cover.

The template's styles.xml, numbering.xml, theme, embedded fonts, headers, footers and page setup are applied to the new content. Everything the merge does not regenerate is carried through byte-for-byte, so the letterhead, logo and licensed fonts come out exactly as they went in.

What comes from where:

Templatepage size, margins, headers, footers, page numbering, fonts, theme, colours
Your documentthe body content, and docProps/core.xml (title/author)
Mergedstyles.xml (ours fill only the styleIds the template lacks), numbering.xml (our list definitions take ids above the template's), [Content_Types].xml, both .rels

Two consequences worth knowing:

  • The template's headers and footers win. A source document's own headers are dropped rather than mixed with the letterhead. The page furniture all lives in the template's body <w:sectPr>, which is lifted whole.
  • The template's comments are dropped along with its body, and so are commentsExtended.xml and people.xml. Comments in the source document still come through.

Two flags refine a multi-section template:

  • --reference-section N picks which of the template's sections supplies the page setup, headers and footers — 1 is the first section, Word's numbering. The default is the last section (the body-level one, what the flag-less behaviour has always lifted). A multi-section template's wanted furniture is often an earlier section's — the master agreement's CONFIDENTIAL footer, not the Annex's missing one.
  • --table-style NAME binds generated tables to a table style the template defines (matched on styleId, then style name). Without it, the style named Table is used if the template has one, else the first table style that is not the implicit Normal Table. Under a bound style the generator stops emitting its own hardcoded borders — the style carries them.

An unreadable or non-DOCX reference is an error and writes nothing — a silent fallback to the built-in styling would produce a plausible file missing exactly the letterhead it was asked for. DOCX output only.

Architecture

docparse/├── types/document.ail           # Block ADT (11 variants)├── services/│   ├── format_router.ail        # Format detection (36 inline tests)│   ├── zip_extract.ail          # ZIP layer (9 inline tests)│   ├── docx_parser.ail          # DOCX XML → Blocks (6 inline tests)│   ├── pptx_parser.ail          # PPTX slides → Blocks│   ├── xlsx_parser.ail          # XLSX worksheets → Blocks│   ├── direct_ai_parser.ail     # PDF/image → Blocks (AI)│   ├── layout_ai.ail            # AI self-healing (optional)│   ├── output_formatter.ail     # JSON + markdown output│   └── docparse_browser.ail     # WASM browser adapter└── main.ail                     # CLI entry point

91 contracts, 50+ inline tests. Of the 91, Z3 proves 14 outright; the rest are checked at runtime under --prove/--verify-contracts in CI, and skip statically because parser code is recursive and higher-order, which is outside Z3's decidable fragment.

AI Configuration

AILANG Parse uses AILANG's AI effect — any model AILANG supports works:

bash
docparse scan.pdf --ai gemini-3.5-flash-lite          # Google (default; fast)docparse scan.pdf --ai granite-docling           # Local Ollama (free)docparse scan.pdf --ai claude-haiku-4-5          # Anthropic

AI usage is bounded by capability budgets (AI @limit=200 on main), so costs are predictable.

Dev Commands

bash
docparse --check       # Type-check all modulesdocparse --test        # Run inline testsdocparse --prove       # Static Z3 contract verification

Benchmarks

bash
uv run benchmarks/run_benchmarks.py --suite office     # Structural (no API, instant)uv run benchmarks/run_benchmarks.py --suite pdf         # PDF extraction (needs AI)uv run benchmarks/run_benchmarks.py --competitors       # Compare to Docling etc.

See benchmarks/ for details.

License

Apache 2.0

來源：README.md，提交 b48b321

工具

10

createUpload

Create upload link

Get one-time upload links for files in your sandbox or workspace, when you can run shell commands with network access (Claude Code, Codex, a code sandbox). One token PER FILE: a token uploads exactly one file. For several files set count (1..20) to get that many descriptors in one call. Each descriptor has a curl line: run it where the file is (replace <path>); it prints a receipt with a fileRef. Then call mcpParse, mcpConvert or editDocument with that fileRef. Tokens work once, for 10 minutes. If curl fails with a network error, or the user wants to pick a file, call uploadDocument instead, or ask for an https:// link. Never parse the file yourself.

輸入參數countfilenamemimeType

editDocument

Edit document

Edit a document: parse it, apply a JSON array of edit deltas, and return the modified blocks (same shape as mcpParse with outputFormat "blocks"). deltas "" or "[]" returns the document unchanged. Office formats only: docx, pptx, xlsx, odt, odp, ods. Give the document ONE way: fileRef: a file the user uploaded (mcp-file://docparse/file_...). file: in ChatGPT, the file the user attached (ChatGPT fills it in). content + filename: the file base64-encoded, contentEncoding="base64", e.g. filename "contract.docx". filepath: an https:// URL of a publicly reachable document, or a sample id from mcpFormats such as "sample_docx_formatting". To get a file back, pass the edited blocks' Markdown to mcpConvert. Which way to send a file: a file in your sandbox or workspace: call createUpload (one token PER FILE; set count for several) and run its curl line, then pass the fileRef here. curl fails with a network error, or the user wants to pick a file: call uploadDocument (the user chooses it in the AILANG Parse card). The card parses the file; it does not convert or edit it. a public https:// link; short text as content. In ChatGPT, file. Never parse the file yourself (your own script, or your own reading of it) and present that as this tool's result: if the file cannot be sent, tell the user.

輸入參數filedeltascontentfileReffilenamefilepathcontentEncoding

mcpAccount

Account, quota and usage

Your AILANG Parse account: plan, monthly usage and limits, your keys, and whether request history is kept. Actions: "status" (default) → plan, usage this month and remaining allowance "keys" → your keys with per-key usage ("usage" is an alias) "history" → whether request history (parsed output kept for replay) is on for your account "history_on" / "history_off" → change it for the account and all its keys

輸入參數action

mcpConvert

Convert document

Create a document in another format: docx, pptx, xlsx, odt, odp, ods, html, md or qmd. To write a new document, write Markdown and send it as content with filename "notes.md" (headings become slides in pptx). To convert an existing document, give it ONE way: fileRef: a file the user uploaded (mcp-file://docparse/file_...). file: in ChatGPT, the file the user attached (ChatGPT fills it in). content + filename: the document itself; binary files base64-encoded with contentEncoding="base64", text as plain text. input: an https:// URL of a publicly reachable document, or a sample id from mcpFormats. Which way to send a file: a file in your sandbox or workspace: call createUpload (one token PER FILE; set count for several) and run its curl line, then pass the fileRef here. curl fails with a network error, or the user wants to pick a file: call uploadDocument (the user chooses it in the AILANG Parse card). The card parses the file; it does not convert or edit it. a public https:// link; short text as content. In ChatGPT, file. Never parse the file yourself (your own script, or your own reading of it) and present that as this tool's result: if the file cannot be sent, tell the user. The result carries download_url, a link to the generated file valid for one hour: give it to the user. It also carries the file as base64 in content (encoding "base64"; text formats come back as utf8). Leave outputPath empty: the server cannot write to your filesystem.

輸入參數fileinputcontentfileReffilenameoutputPathoutputFormat必填contentEncoding

mcpEstimate

Estimate cost and latency

Estimate cost and latency for parsing a document. Accepts a file path or sample_id. No auth required.

輸入參數filepath必填outputFormat必填

mcpFormats

List formats and capabilities

What AILANG Parse can read and write, how to send it a document, and the sample documents you can try without one. No sign-in needed.

無需輸入參數。

mcpParse

Parse document

Parse a document into structured blocks, Markdown, HTML or A2UI, or a short summary of it. Give the document ONE way: fileRef: an uploaded file (mcp-file://docparse/file_...). The same user can use it again for 30 minutes after the first use (more parts, another format), then it is deleted. file: in ChatGPT, the file the user attached (ChatGPT fills it in). filepath: an https:// URL of a publicly reachable document, or a sample id from mcpFormats such as "sample_docx_formatting". content + filename: short text only (Markdown, HTML, CSV); the filename's extension picks the parser. outputFormat: blocks (default), markdown, html, a2ui, or summary (counts, the first 50 headings, warnings: a few KB). Look with summary first when a document may be large. Large documents come in parts: when the result has page.nextCursor, call mcpParse again with the same fileRef/filepath and cursor=<nextCursor> for the next part; read every part before answering about the whole document. maxChars sets the part size as a string of digits (default "80000", "2000".."120000"). pdfBackend: leave empty for PDFs with a text layer. Set "ai" when the result is PDF_TEXT_LAYER_SPARSE (a scanned PDF): each page is read by an AI model, about 5 seconds a page, and the call counts as an AI request. Pass the same pdfBackend with every cursor call. Which way to send a file: a file in your sandbox or workspace: call createUpload (one token PER FILE; set count for several) and run its curl line, then call this tool with the fileRef the upload prints. curl fails with a network error, or the user wants to pick a file: call uploadDocument; the user chooses it in the AILANG Parse card, which uploads it and gives you its fileRef. a public https:// link: filepath. Short text: content. Never parse the file yourself (your own script, or your own reading of it) and present that as this tool's result: if the file cannot be sent, tell the user.

輸入參數filecursorcontentfileReffilenamefilepathmaxCharsrequestIdpdfBackendoutputFormatcontentEncoding

submit_feedback

Send feedback

Anonymous bug report / feature request / docs gap, queued for human review. Default routing: `public-feedback` inbox (general AILANG). Pass `package="vendor/name"` (e.g. "sunholo/auth") to route to that package's `pkg:vendor/name` inbox where its autonomous agent watches. Categories: bug, feature, docs, limitation. Body limit 10KB, snippet limit 4KB. Optional contact field for follow-up; opaque to the server. Set `auto_dispatch=true` to authorize the package agent to act on your submission immediately (default false — files for human triage; pkg-feedback agent template lands in a separate sprint).

輸入參數body必填title必填contactpackagesnippetcategory必填auto_dispatchailang_version必填

uploadDocument

Upload a document

Let the user pick a file to parse: shows the AILANG Parse card with a Choose file button. The card uploads the file and tries to summarise it for you; either way you get its fileRef. Then call mcpParse with that fileRef for the content (the same user can reuse it for 30 minutes; a large document comes in parts). Use this when the user wants to pick a file, or when createUpload's curl fails with a network error. For a file already in your sandbox or workspace, use createUpload + curl instead. Never parse the file yourself.

輸入參數outputFormat

uploadViaHost

Receive upload from widget

Receive a file from the AILANG Parse upload widget, relayed by the host as base64. Only the widget calls this; ticket is the upload token it holds.

輸入參數base64必填ticket必填filename必填

版本歷史

3
  1. v0.14.1最新Nov 5, 2026
  2. v0.14.0Nov 3, 2026
  3. v0.13.1Sep 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

其他 Parse 整合

Parse

gl.parse官方發布者

讓助理查詢 AI 引擎如何提及、推薦與引用品牌、主題和問題，並管理隨之產生的後續任務。

Streamable HTTP可網頁執行
效率與工作流30 個工具Oct 24, 2026

PdfParse

由 GitHub 帳號 PabloG6 發布

託管型 MCP 伺服器，可將 PDF 轉成結構化表格、讓助理查詢擷取出的紀錄，並設定文件收件匣工作流程。

Streamable HTTP可網頁執行驗證方式
檔案與儲存需要憑證Oct 24, 2026

DocParse

由 GitHub 帳號 mrsergeiluzhnov 發布

DocParse 是遠端 MCP 伺服器，可將文件轉為 Markdown，並提供美國金融資料（SEC 申報、公債殖利率、FRED 總經、匯率）及若干開發小工具。

Streamable HTTP可網頁執行
檔案與儲存12 個工具Oct 24, 2026

SemVer numbers only

由 GitHub 帳號 sadri-dridi 發布

一個遠端 MCP 端點，提供 31 個小型工具：SemVer 解析、時間與時區查詢、URL 與主機名稱檢查、HTTP 狀態檢查以及簡單的格式驗證。

Streamable HTTP可網頁執行
網頁搜尋與抓取31 個工具Nov 2, 2026

ParseRelay

由 GitHub 帳號 parserelay 發布

ParseRelay 的 MCP 伺服器提供單一 scan 工具，將文件解析成附信心分數的結構化欄位，供各種 MCP 用戶端使用。

STDIO僅桌面
檔案與儲存尚未驗證Oct 24, 2026