MinerU PDF Parser

用 MinerU API 解析 PDF/Word/PPT/图片为 Markdown，支持公式、表格、OCR。适用于论文解析、文档提取。

Security Scan

The skill appears to do what it says (call MinerU to parse documents) but the SKILL.md expects an API key and command-line tools that the registry metadata does not declare — this mismatch is suspicious and should be resolved before installing.

Purpose & Capability

The name/description match the instructions: the skill delegates PDF/Word/PPT/image → Markdown conversion to MinerU's API. However, SKILL.md demonstrates use of an API key (MINERU_TOKEN) and uses MinerU endpoints while the registry metadata declares no required environment variables or primary credential. That omission is inconsistent with the skill's stated purpose.

Instruction Scope

Runtime instructions include network calls to https://mineru.net, uploading files via presigned URLs, polling task status, and local commands (curl, jq, unzip, mkdir). Those actions are coherent with document parsing, but the instructions reference environment variables and CLI tools that are not declared in the registry. The skill also suggests adding MINERU_TOKEN to ~/.bashrc — i.e., storing an API key locally — which is normal for API use but should be explicitly declared.

Install Mechanism

This is an instruction-only skill with no install spec and no code files, which is the lowest-risk install pattern. Nothing is written to disk by the skill bundle itself.

Credentials

SKILL.md requires an API credential (MINERU_TOKEN) for MinerU; the registry metadata lists no required environment variables or primary credential. Requesting a single service API key is proportionate to the task, but the lack of declaration is an inconsistency that can hide how secrets will be used. No unrelated credentials are requested.

Persistence & Privilege

The skill does not request 'always: true' or other elevated persistence and does not claim to modify other skills or system-wide settings. Autonomous invocation is allowed (platform default) and appropriate for a callable integration.

What to consider before installing

Before installing, note these points and take these steps: - The SKILL.md shows the skill uses an API key named MINERU_TOKEN (Authorization: Bearer) and makes network calls to mineru.net. The registry metadata did NOT declare any required environment variable — ask the publisher to correct the metadata so you can see exactly what secrets are needed. - The instructions rely on command-line tools (curl, jq, unzip, mkdir). Ensure those tools are available and that the agent environment will handle them safely. - Consider the sensitivity of documents you will upload. MinerU is an external service; uploading private/confidential docs will transmit them off your machine. If you need to protect data, do not provide highly sensitive files or create a limited-scope API key. - Verify the skill origin: source is listed as unknown. Check the MinerU homepage and API docs (links exist in SKILL.md) and prefer installing skills from identifiable publishers or from the project's official GitHub (links are provided). Ask the publisher to add the required env var(s) to the registry entry and to confirm whether any data persists on mineru.net. - If you proceed, create and use an API key with minimal scope/quota, and rotate or revoke it after testing. Refuse to provide unrelated credentials (AWS, GitHub tokens, etc.) — they are not needed for this skill. If the publisher cannot clarify the missing metadata (required env var and expected CLI dependencies), treat the skill as untrusted and avoid installing it.

📄 MinerU - 文档解析神器

OpenDataLab 出品

PDF/Word/PPT/图片 → 结构化 Markdown，公式表格全保留！

---

🔗 资源链接

资源	链接
官网	https://mineru.net/
API 文档	https://mineru.net/apiManage/docs
GitHub	https://github.com/opendatalab/MinerU

---

🎯 功能

支持的文件类型

类型	格式
📕 PDF	论文、书籍、扫描件
📝 Word	.docx
📊 PPT	.pptx
🖼️ 图片	.jpg, .png (OCR)

核心优势

1. 公式完美保留 - LaTeX 格式输出
2. 表格结构识别 - 复杂表格也能搞定
3. 多语言 OCR - 中英文混排无压力
4. 版面分析 - 多栏、图文混排自动处理

---

🚀 API 使用 (v4)

认证

# Header 认证
Authorization: Bearer {YOUR_API_KEY}

单文件解析

# 1. 提交任务
curl -X POST "https://mineru.net/api/v4/extract/task" \
  -H "Authorization: Bearer $MINERU_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "url": "https://arxiv.org/pdf/2410.17247",
    "enable_formula": true,
    "enable_table": true,
    "layout_model": "doclayout_yolo",
    "language": "en"
  }'

# 返回: {"task_id": "xxx", "status": "pending"}

# 2. 轮询结果
curl "https://mineru.net/api/v4/extract/task/{task_id}" \
  -H "Authorization: Bearer $MINERU_TOKEN"

# 返回: {"status": "done", "result": {...}}

批量解析

# 1. 获取上传 URL
curl -X POST "https://mineru.net/api/v4/file-urls/batch" \
  -H "Authorization: Bearer $MINERU_TOKEN" \
  -d '{"file_names": ["paper1.pdf", "paper2.pdf"]}'

# 2. 上传文件到返回的 presigned URLs

# 3. 批量提交任务
curl -X POST "https://mineru.net/api/v4/extract/task/batch" \
  -H "Authorization: Bearer $MINERU_TOKEN" \
  -d '{"files": [{"url": "...", "name": "paper1.pdf"}, ...]}'

---

⚙️ 参数说明

参数	类型	说明
url	string	文件 URL (支持 http/https)
enable_formula	bool	启用公式识别 (默认 true)
enable_table	bool	启用表格识别 (默认 true)
layout_model	string	doclayout_yolo (快) / layoutlmv3 (准)
language	string	en / ch / auto
model_version	string	pipeline / vlm / MinerU-HTML

模型版本对比

版本	速度	准确度	适用场景
pipeline	⚡ 快	高	常规文档
vlm	🐢 慢	最高	复杂版面
MinerU-HTML	⚡ 快	高	网页样式输出

---

📂 输出结构

解析完成后下载的 ZIP 包含：

output/
├── full.md           # 完整 Markdown
├── content_list.json # 结构化内容
├── images/           # 提取的图片
└── layout.json       # 版面分析结果

---

🔧 OpenClaw 集成工作流

论文解析流程

# 1. 创建论文目录
mkdir -p "./paper-reading/[CVPR 2025] NewPaper"
cd "./paper-reading/[CVPR 2025] NewPaper"

# 2. 提交解析任务
TASK_ID=$(curl -s -X POST "https://mineru.net/api/v4/extract/task" \
  -H "Authorization: Bearer $MINERU_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"url": "https://arxiv.org/pdf/XXXX.XXXXX"}' | jq -r '.task_id')

# 3. 等待完成 & 下载
# (轮询 status 直到 done，然后下载 result.zip)

# 4. 解压
unzip result.zip -d .

环境变量

在 ~/.bashrc 或 OpenClaw config 中设置：

export MINERU_TOKEN="your_api_key_here"

---

⚠️ 限制

限制	数值
单文件大小	200 MB
单文件页数	600 页
并发任务数	根据套餐

---

💡 使用技巧

1. arXiv 论文直接用 URL

   https://arxiv.org/pdf/2410.17247
   

2. 中文论文用 language: ch

3. 复杂表格用 vlm 模型

4. 批量处理省 quota

   • 一次提交多个文件，比单个提交更高效

---

📚 相关资源

• Paper Banana Skill - 论文配图生成

---

论文解析不再手动复制粘贴！📖