본문으로 건너뛰기

MinerU PDF Parsing

Purpose​

Call the MinerU service to parse PDF documents, extracting high-quality Markdown text and images, producing directly readable note files.

MinerU is a deep learning-based PDF parsing tool that extracts high-quality text and figures from academic papers.

Use Cases​

  • Converting PDF-format literature into editable Markdown
  • Preparing plain text documents for downstream workflows (e.g., Literature Analysis, Deep Reading)
  • Extracting images and tables from PDFs

Configuring the MinerU Backend​

1. Register a MinerU Account and Obtain an API Token​

  1. Visit mineru.net to register an account
  2. After logging in, go to the API → API Management page
  3. Create or copy an API Token

2. Add a Backend in Backend Manager​

  1. Open Tools → Backend Manager
  2. Switch to the Generic HTTP tab
  3. Click Add Generic HTTP
  4. Fill in the following fields:
FieldValue
Display NameMinerU Official (or any name you prefer)
Base URLhttps://mineru.net
Auth Methodbearer
Auth TokenPaste the API Token obtained in the previous step
Timeout600000 (10 minutes)
  1. Click Save in the bottom-right corner

Input Constraints​

Constraint TypeDescription
Input UnitAttachment
Accepted Typesapplication/pdf (PDF only)
Conflict DetectionIf a .md file with the same name already exists in the same directory, the PDF is skipped

Trigger Methods​

  • Directly select one or more PDF attachments
  • Select the parent item, and the plugin will automatically expand its child PDF attachments

Conflict Handling​

  • Checks whether <PDF filename>.md exists in the target directory
  • If it exists, the input is skipped during preprocessing
  • If all candidates have conflicts, the workflow does not submit any tasks

Execution Flow​

1. Request Upload URL
   └── POST to MinerU API to obtain batch_id and upload_url

2. Upload File
   └── Binary upload of the PDF file

3. Poll for Results
   └── Repeated queries until processing completes or fails
       └── Interval: 2 seconds

4. Download Results
   └── Download bundle (zip format)

5. Local Materialization
   └── Extract bundle
       └── Extract Markdown content
       └── Extract images
       └── Rewrite image paths in Markdown to local relative paths
       └── Write to the same directory as the PDF

Outputs​

1. Markdown File​

  • Location: Same directory as the PDF
  • Naming: <original filename>.md
  • Content: Parsed Markdown text
  • Encoding: UTF-8

2. Image Directory​

  • Location: Same directory as the PDF: Images_<ItemKey>/
  • Content: Image files extracted from the PDF

3. Linked Attachment​

  • Type: Link to local file
  • Location: Under the parent item
  • Target: The .md file

Cleanup Logic​

  • If Images_<ItemKey>/ already exists in the target directory, the old directory is deleted before writing
  • Avoids creating duplicate .md linked attachments that already exist

Estimated Duration​

PDF SizeEstimated Time
Short paper (≤15 pages)30 seconds - 1 minute
Standard (15-40 pages)1-2 minutes
Long paper (40+ pages)2-3 minutes

Duration mainly depends on the processing speed of the MinerU service.

Parameters​

The MinerU workflow has no user-configurable parameters.

Model Recommendation​

No LLM model required. This workflow only calls the MinerU service via HTTP API.

Dependencies​

  • Backend: MinerU service (Generic HTTP backend)
  • Backend Configuration: Configure a Generic HTTP type backend in Backend Manager
  • Authentication: A valid API Token (Bearer token) is required
  • MinerU Service URL: https://mineru.net or another deployed instance