MinerU Open MCP — Claude Code MCP server for document parsing
Turns PDFs, images and Office files into Markdown from inside your agent chat. Useful when you need to feed a report or paper to the model.
claude mcp add --env MINERU_API_TOKEN=<MINERU_API_TOKEN> --env OUTPUT_DIR=<OUTPUT_DIR> --transport stdio mineru-open-mcp -- uvx mineru-open-mcpWhat MinerU Open MCP does
MinerU Open MCP is a Python server that exposes the MinerU Open API to your agent. Through it, Claude, Codex, Cursor and other clients read PDF, images, doc, docx, ppt, pptx, xls, xlsx and html, and get Markdown back. Formulas come out as LaTeX, tables as HTML, headers and footers are dropped, and reading order is preserved.
There are two modes. Precision Extract needs a token, accepts up to 200 MB and 200 pages, supports batch runs and returns Markdown, JSON or Zip with optional DOCX, HTML and LaTeX export. Quick Parse skips auth but caps at 10 MB and 20 pages, Markdown only, one file at a time. The token goes in MINERU_API_TOKEN, the output folder in OUTPUT_DIR.
It runs through uvx, so uv has to be installed first. Without a token the server still starts, but it hits the quick-mode limits — worth checking which path you actually need. For large scans and batch work, go with Precision Extract.
What it helps with
- pull text out of a PDF with tricky layout and tables
- convert scans and handwritten pages to Markdown
- capture formulas from a research paper as LaTeX
- process a batch of Office files in one pass
- hand a web page's content to the agent without copy-paste
Example prompts for the agent
- Parse
report.pdfand save the Markdown to./output/ - Extract the tables from
paper.pdfand export them as HTML - Crawl this page and show me the text as Markdown
How to install MinerU Open MCP
Platform CLI
claude mcp add --env MINERU_API_TOKEN=<MINERU_API_TOKEN> --env OUTPUT_DIR=<OUTPUT_DIR> --transport stdio mineru-open-mcp -- uvx mineru-open-mcpReplace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from mineru-open-mcp <command>
Platform CLI
codex mcp add mineru-open-mcp --env MINERU_API_TOKEN=<MINERU_API_TOKEN> --env OUTPUT_DIR=<OUTPUT_DIR> -- uvx mineru-open-mcpReplace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from mineru-open-mcp <command>
Config file
{
"mcpServers": {
"mineru-open-mcp": {
"command": "uvx",
"args": [
"mineru-open-mcp"
],
"env": {
"MINERU_API_TOKEN": "<MINERU_API_TOKEN>",
"OUTPUT_DIR": "<OUTPUT_DIR>"
}
}
}
}Replace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from mineru-open-mcp <command>
Config file
{
"servers": {
"mineru-open-mcp": {
"type": "stdio",
"command": "uvx",
"args": [
"mineru-open-mcp"
],
"env": {
"MINERU_API_TOKEN": "<MINERU_API_TOKEN>",
"OUTPUT_DIR": "<OUTPUT_DIR>"
}
}
}
}Replace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from mineru-open-mcp <command>
Platform CLI
gemini mcp add -e MINERU_API_TOKEN=<MINERU_API_TOKEN> -e OUTPUT_DIR=<OUTPUT_DIR> mineru-open-mcp uvx mineru-open-mcpReplace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from mineru-open-mcp <command>
Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about MinerU Open MCP
How do I install MinerU Open MCP in Claude?
Run claude mcp add mineru-open-mcp -e MINERU_API_TOKEN=<MINERU_API_TOKEN> -- uvx mineru-open-mcp. For a custom output folder, use the variant with --env MINERU_API_TOKEN=<MINERU_API_TOKEN> --env OUTPUT_DIR=<OUTPUT_DIR> --transport stdio.
What does MinerU Open MCP do?
It gives your agent a document-parsing tool backed by the MinerU Open API: PDFs, images, Office files and HTML become Markdown or JSON.
How is Precision Extract different from Quick Parse?
Precision Extract requires a token, takes files up to 200 MB and 200 pages, supports batch processing and several output formats. Quick Parse needs no auth but is limited to 10 MB, 20 pages and a single file.
