Skip to main content
skill CX Codex

deepseek-ocr — Claude Code skill: what it does, how to install

Teaches your coding agent to run DeepSeek-OCR for document parsing and markdown extraction. Works with images and PDFs via vLLM or Transformers.

Install into Claude Code · Skills CLI (Vercel) All methods and platforms ↓
npx skills add reason-machines/trending-skills --skill deepseek-ocr -a claude-code
Open on GitHub

What deepseek-ocr does

This skill guides coding agents through configuring DeepSeek-OCR pipelines. The model processes visual document inputs and outputs structured markdown or plain text. It covers dedicated prompt formats for tabular data extraction, bounding box detection via grounding tokens, and chart interpretation.

Setups can use HuggingFace Transformers for isolated runs or vLLM when batch performance matters. The instructions include logit processing configurations with NGramPerReqLogitsProcessor to preserve HTML table tags. Execution fails without CUDA 11.8+ and PyTorch 2.6.0.

What it helps with

  • convert scanned multi-page PDFs into formatted markdown
  • extract plain text from document pictures without layout tags
  • parse diagrams and visual charts into text
  • detect bounding coordinates for target words on a page

Example prompts for the agent

  • Convert contract.pdf to markdown using the DeepSeek-OCR vLLM script
  • Write a HuggingFace inference script to parse chart.png
  • Extract all text from invoice.jpg without layout structure

How to install deepseek-ocr

1Skills CLI (Vercel)

npx skills add reason-machines/trending-skills --skill deepseek-ocr -a claude-code

2Manually: put the files in place

~/.claude/skills/deepseek-ocr/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/deepseek-ocr/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about deepseek-ocr

How to install deepseek-ocr in Claude Code?

Run npx skills add reason-machines/trending-skills --skill deepseek-ocr -a claude-code in your terminal.

What does deepseek-ocr do?

It supplies the agent with prompt templates and execution patterns to run DeepSeek-OCR on PDFs and images.

What dependencies are needed to run the model?

The setup requires Python 3.12.9, PyTorch 2.6.0, and CUDA 11.8+. You also need either vLLM 0.8.5 or HuggingFace Transformers alongside flash-attn==2.7.3.

Similar skills

Everything for AI and ML →

From the same set49

inkos-multi-agent-novel-writing Coordinates a multi-agent CLI pipeline to draft, validate, and audit novel chapters. Useful for managing long-form fiction projects with strict continuity. CX Codex +1 2,800installs daily-stock-analysis Automates stock analysis across US, Hong Kong, and Chinese markets using LLMs. It calculates entry targets, stop-losses, and delivers reports via chat webhooks. CX Codex +1 2,000installs paperclip-ai-orchestration Teaches coding assistants to manage Paperclip, an open-source platform that runs autonomous AI agents with org charts, task tickets, and budget caps. CX Codex +1 1,500installs lightpanda-browser A skill that teaches your agent to run Lightpanda: install the Zig headless browser, start a CDP server, and hook Playwright or Puppeteer to it. CX Codex +1 1,500installs vinext-vite-nextjs Guides your AI assistant through running Next.js codebases on Vite and deploying them to Cloudflare Workers. CX Codex +1 1,400installs minecraftconsoles-lce This skill guides you through compiling, running, and modding the C++ Minecraft Legacy Console Edition project. It covers CMake builds, dedicated servers, and controls setup. CX Codex +1 1,400installs open-autoglm-phone-agent A skill for Open-AutoGLM that walks your coding agent through deploying the model and driving Android, HarmonyOS NEXT or iOS devices. CX Codex +1 1,400installs voicebox-voice-synthesis Tells your agent how to run Voicebox and talk to its REST API on localhost:17493: speech generation, voice profiles, effects, queue. CX Codex +1 1,400installs

Project sponsors

Companies that keep this analytics open
Gold sponsor

Impressions, clicks and code copies — open statistics. We keep the numbers public for the same reason NeBlask itself is public. Become a sponsor