MinerU Open MCP

MinerU-Ecosystem

The official ecosystem toolkit for [MinerU](https://github.com/opendatalab/MinerU) Open API

Empowering developers and AI agents with seamless document parsing capabilities — PDF · Word · PPT · Images · Web pages → Markdown / JSON · VLM+OCR dual engine · 109 languages · MCP Server · LangChain / RAGFlow / Dify / FastGPT native integration.

中文文档

---

📖 Overview

MinerU-Ecosystem provides a full suite of tools, SDKs, and integrations built on top of the MinerU Open API. Whether you're building production pipelines, integrating with LangChain for RAG, or enabling AI agents to parse documents on the fly — this repository has you covered.

MinerU is an open-source, high-accuracy document parsing engine that converts unstructured documents (PDFs, images, Office files, etc.) into machine-readable Markdown and JSON, purpose-built for LLM pre-training, RAG, and agentic workflows.

Core capabilities:

• Formulas → LaTeX · Tables → HTML, accurate complex layout reconstruction
• Supports scanned docs, handwriting, multi-column layouts, cross-page table merging
• Output follows human reading order with automatic header/footer removal
• VLM + OCR dual engine, 109-language OCR recognition

---

🏗️ Repository Structure

MinerU-Ecosystem/

├── cli/ # Command-line tool for document parsing ├── sdk/ # Multi-language SDKs │ ├── python/ # Python SDK │ ├── go/ # Go SDK │ └── typescript/ # TypeScript SDK ├── langchain_mineru/ # LangChain document loader integration ├── llama-index-readers-mineru/ # LlamaIndex document reader integration ├── mcp/ # Model Context Protocol server (Python) └── skills/ # AI agent skills (Claude Code, OpenClaw, etc.)

---

🔑 Supported APIs

All components support both API modes:

🧭 Choose Your Integration Path

Not sure where to start? Pick the path that matches your use case:

I want to...

│ ├── 🌐 Try it instantly, with no install and no code │ └── Web App → https://mineru.net/OpenSourceTools/Extractor │ ├── 💻 Parse documents from the terminal │ └── CLI → cli/ │ flash-extract: no token, best for quick previews │ extract: full features, better for production workflows │ ├── 🐍 Integrate it into my Python / Go / TypeScript project │ └── SDK → sdk/python/ | sdk/go/ | sdk/typescript/ │ ├── 🤖 Enable my AI agent to parse documents │ ├── Call the CLI directly → cli/ │ ├── Use natural-language skills (OpenClaw, ZeroClaw, etc.) → skills/ │ └── Use MCP protocol (Cursor, Claude Desktop, Windsurf, etc.) → mcp/ │ ├── 📚 Build a RAG pipeline / knowledge base │ ├── LangChain Loader → langchain_mineru/ │ └── LlamaIndex Reader → llama-index-readers-mineru/ │ flash mode: zero-token quick start │ precision mode: OCR, tables, formulas, and higher fidelity

---

🚀 Quick Start

💻 CLI (cli/)

A fast command-line tool for parsing documents directly from your terminal.

Installation

# Linux / macOS

curl -fsSL https://cdn-mineru.openxlab.org.cn/open-api-cli/install.sh | sh

# Windows (PowerShell)

irm https://cdn-mineru.openxlab.org.cn/open-api-cli/install.ps1 | iex

Flash Extract (no login)

mineru-open-api flash-extract report.pdf

Precision Extract (login required)

# First-time setup

mineru-open-api auth

Extract to stdout

mineru-open-api extract paper.pdf

Save all resources (images/tables) to directory

mineru-open-api extract report.pdf -o ./output/

Export to multiple formats

mineru-open-api extract report.pdf -f docx,latex,html -o ./results/

Web Crawl

mineru-open-api crawl https://www.example.com

Batch Processing

# All PDFs in current directory

mineru-open-api extract *.pdf -o ./results/

From a file list

mineru-open-api extract --list filelist.txt -o ./results/

---

🐍 Python SDK

Installation

pip install mineru-open-sdk

Flash Extract (no token)

from mineru import MinerU

client = MinerU() result = client.flash_extract("https://cdn-mineru.openxlab.org.cn/demo/example.pdf") print(result.markdown)

Precision Extract (token required)

from mineru import MinerU

client = MinerU("your-api-token") result = client.extract("https://cdn-mineru.openxlab.org.cn/demo/example.pdf") print(result.markdown) print(result.images) # extracted image list

---

🐹 Go SDK

Installation

go get github.com/opendatalab/MinerU-Ecosystem/sdk/go@latest

Flash Extract

package main

import ( "context" "fmt" mineru "github.com/opendatalab/MinerU-Ecosystem/sdk/go" )

func main() { client := mineru.NewFlash() result, err := client.FlashExtract( context.Background(), "https://cdn-mineru.openxlab.org.cn/demo/example.pdf", ) if err != nil { panic(err) } fmt.Println(result.Markdown) }

Precision Extract

client, err := mineru.New("your-api-token")

if err != nil { panic(err) } result, err := client.Extract( context.Background(), "https://cdn-mineru.openxlab.org.cn/demo/example.pdf", ) if err != nil { panic(err) } fmt.Println(result.Markdown)

Precision Extract with options

result, err := client.Extract(ctx, "./paper.pdf",

mineru.WithModel("vlm"), mineru.WithLanguage("en"), mineru.WithPages("1-20"), mineru.WithExtraFormats("docx"), mineru.WithPollTimeout(10*time.Minute), ) if err != nil { panic(err) } if err := result.SaveAll("./output"); err != nil { panic(err) }

Batch Processing

ch, err := client.ExtractBatch(ctx, []string{"a.pdf", "b.pdf"})

if err != nil { panic(err) } for result := range ch { fmt.Printf("%s: %s\n", result.Filename, result.State) }

Web Crawling

result, err := client.Crawl(ctx, "https://www.example.com")

if err != nil { panic(err) } fmt.Println(result.Markdown)

---

🟦 TypeScript / JavaScript SDK

Installation

npm install mineru-open-sdk

Flash Extract

import { MinerU } from "mineru-open-sdk";

const client = new MinerU(); const result = await client.flashExtract( "https://cdn-mineru.openxlab.org.cn/demo/example.pdf" ); console.log(result.markdown);

Precision Extract

import { MinerU } from "mineru-open-sdk";

const client = new MinerU("your-api-token"); const result = await client.extract( "https://cdn-mineru.openxlab.org.cn/demo/example.pdf" ); console.log(result.markdown); console.log(result.images);

Precision Extract with options

import { MinerU, saveAll } from "mineru-open-sdk";

const client = new MinerU("your-api-token"); const result = await client.extract("./paper.pdf", { model: "vlm", // "vlm" | "pipeline" | "html" language: "en", pages: "1-20", extraFormats: ["docx"], timeout: 600, }); await saveAll(result, "./output");

Batch Processing

for await (const result of client.extractBatch(["a.pdf", "b.pdf"])) {

console.log(${result.filename}: ${result.state}); }

Web Crawling

const result = await client.crawl("https://www.example.com");

console.log(result.markdown);

---

🤖 Use with Claude / Cursor (MCP Server)

MinerU provides an official MCP Server allowing Claude Desktop, Cursor, Windsurf, and any MCP-compatible AI client to parse documents as a native tool.

No API key needed — Flash mode works out of the box, free, up to 20 pages / 10 MB per file.

Configure: `claude_desktop_config.json` / `.cursor/mcp.json`

{

"mcpServers": { "mineru": { "command": "uvx", "args": ["mineru-open-mcp"], "env": { "MINERU_API_TOKEN": "your_key_here" } } } }

Streamable HTTP mode (web-based MCP clients)

MINERU_API_TOKEN=your_key mineru-open-mcp --transport streamable-http --port 8001

{

"mcpServers": { "mineru": { "type": "streamableHttp", "url": "http://127.0.0.1:8001/mcp" } } }

Tools exposed via MCP:

Environment Variables:

---

🦜 Use in RAG with LangChain

langchain-mineru is an official LangChain Document Loader — parse any document into LangChain Document objects with one line of code.

Installation

pip install langchain-mineru

Minimal example (no token)

1. Basic usage (`flash` mode by default, no token required)

from langchain_mineru import MinerULoader

loader = MinerULoader(source="demo.pdf") # flash mode, no token needed docs = loader.load() print(docs[0].page_content[:500]) print(docs[0].metadata)

Default is mode="flash", which is ideal for quick previews and lightweight integrations.

2. Precision mode (token required)

Best for long documents, larger files, and workflows that need higher-fidelity extraction or standard API outputs. Flash mode also supports OCR, table, and formula switches within flash API limits.

from langchain_mineru import MinerULoader

loader = MinerULoader( source="/path/to/manual.pdf", mode="precision", token="your-api-token", # or set MINERU_TOKEN split_pages=True, pages="1-5", )

docs = loader.load() for doc in docs: print(doc.metadata.get("page"), doc.page_content[:200])

3. Use it in a LangChain RAG pipeline

from langchain_mineru import MinerULoader

from langchain_text_splitters import RecursiveCharacterTextSplitter from langchain_openai import OpenAIEmbeddings from langchain_community.vectorstores import FAISS

loader = MinerULoader(source="demo.pdf", split_pages=True) docs = loader.load()

splitter = RecursiveCharacterTextSplitter(chunk_size=1200, chunk_overlap=200) chunks = splitter.split_documents(docs)

vs = FAISS.from_documents(chunks, OpenAIEmbeddings()) results = vs.similarity_search("What are the key conclusions in this document?", k=3)

for r in results: print(r.page_cont