Doc2X Docs

English

简体中文

English

简体中文

Theme

Doc2X MCP: Gemini CLI

Back: Doc2X MCP Guide

What You Need

  • Node.js (LTS recommended) and npx
  • DOC2X_API_KEY (looks like YOUR_DOC2X_API_KEY)
  • (Optional) other Doc2X MCP config (for example DOC2X_BASE_URL / timeouts / polling): see “Environment Variables” in the Doc2X MCP Guide
  • (Optional) proxy env (HTTPS_PROXY / HTTP_PROXY)

Add MCP Configuration

Run in your terminal:

shell
gemini mcp add --transport stdio -e DOC2X_API_KEY=YOUR_DOC2X_API_KEY doc2x npx -y @noedgeai-org/doc2x-mcp@0.1.5

Check MCP

Open the MCP list and confirm the doc2x server is available (you can expand it to see the tool list).

Check MCP

Use doc2x-mcp

Try a small file first:

  • Prepare ./paper.pdf (or replace with your file path)
  • In chat, explicitly ask to parse and export using the doc2x MCP server

Example prompts:

  • Use doc2x-mcp to parse ./paper.pdf and export the result to the current directory.
  • Use doc2x-mcp to parse ./paper.pdf and tell me the exported file name/path.

Parse Document

Parsing

View Exported Result

Troubleshooting

  • If the server doesn’t show up, ensure gemini mcp add ... succeeded and restart/refresh the session.
  • If startup hangs/fails, verify npx -y @noedgeai-org/doc2x-mcp@0.1.5 works in your terminal first.

Version, paths, and completion checks

This page configures MCP 0.1.5. Existing screenshots illustrate client steps; labels may change with client versions. Use Node.js 18+ and obtain an API Key from the Open Platform. YOUR_DOC2X_API_KEY is a placeholder: keep the real value in private local configuration, not chat or version control.

After connection, confirm the tool list includes doc2x_parse_pdf_submit and export/download tools. Supply an absolute input path and output directory, authorize one submission, retain uid, and wait for parsing success before export. Follow the complete tool sequence and prompt.

Ask for actual saved paths, then open the result and check images, formulas, and page order. A progress screenshot is not export-success evidence. On timeout query the original uid. Input paths must be readable on the MCP host and the output directory writable. Real parsing consumes platform quota; a connected server alone is not task authorization.