Agent Skills › superlinked/sie › parse-document

parse-document

GitHub

通过 Superlinked MCP 将 PDF、扫描件或办公文件转换为 Markdown，避免直接读取原文档占用上下文。适用于用户请求解析、OCR、提取内容、总结或基于文档问答的场景。

packages/sie_mcp/plugin/claude-code/parse-document/SKILL.md superlinked/sie

触发场景

用户要求阅读文档用户要求解析文档用户要求进行 OCR 识别用户要求从文档中提取信息用户要求总结文档内容用户要求回答关于文档的问题

安装

npx skills add superlinked/sie --skill parse-document -g -y

更多选项

非标准路径

npx skills add https://github.com/superlinked/sie/tree/main/packages/sie_mcp/plugin/claude-code/parse-document -g -y

不安装直接使用

npx skills use superlinked/sie@parse-document

指定 Agent (Claude Code)

npx skills add superlinked/sie --skill parse-document -a claude-code -g -y

安装 repo 全部 skill

npx skills add superlinked/sie --all -g -y

预览 repo 内 skill

npx skills add superlinked/sie --list

SKILL.md

Frontmatter

{
    "name": "parse-document",
    "description": "Convert a PDF, scan, image of a page, or office file to clean markdown through the connected Superlinked MCP edge, so the source document is not read into model context directly. Use when the user asks to read, parse, OCR, extract from, summarize, or answer questions about a document."
}

Parse a Document with Superlinked MCP

Use the connected Superlinked MCP server to convert the source file to markdown through the MCP docs_to_markdown tool, then work from the markdown artifact.

Steps

Do not open, view, attach, or read the source document into context. Read its raw bytes only and base64-encode those bytes.
Call the Superlinked MCP docs_to_markdown tool with:
- document_base64: the base64-encoded source bytes
- filename: the original filename
- engine: auto unless the user explicitly asks to force docling or vl-ocr
- ocr: leave false unless forcing the docling OCR path
Write the returned markdown to processed/<source-stem>.md. Create processed/ if it does not exist.
Your next message must begin with a short receipt, then answer from the artifact selectively:
```
Parsed <source> -> processed/<source-stem>.md (<chars> chars)
```
If the MCP metadata includes token-saving figures, include those figures directly under the receipt. Keep estimates marked as estimates.
Answer the user's question from the artifact. Grep or read specific sections; do not read the whole markdown file unless it is small.

Errors

If the MCP call fails, relay the tool error and do not retry blindly. For auth or connector errors, tell the user to reinstall the generated MCP plugin pack or rerun the generated claude mcp add ... --header 'Authorization: Bearer ...' command from INSTALL.md.