paddleocr-doc-parsing · freestylefly — Claude Code skill: parse documents
Parses PDFs and images through PaddleOCR and returns text, tables, formulas, charts and reading order. Useful when you need the structure of a report or paper, not just raw text.
npx skills add freestylefly/canghe-skills --skill paddleocr-doc-parsing -a claude-codeWhat paddleocr-doc-parsing does
The skill calls the PaddleOCR Document Parsing API through python scripts/vl_caller.py. Pass a link with --file-url or a local file with --file-path; --file-type sets the input kind (0 for PDF, 1 for image). By default the raw JSON lands in a temp file and the script prints the path to stderr. Add --stdout if you want the JSON in the console and no file written.
The response is an envelope with a text field plus the raw result. Per-page data lives in result[n].markdown and result[n].prunedResult, which carries layout, content and confidence. The agent then shows what was asked for: the full text, every table, formulas as LaTeX.
Before the first run you need PADDLEOCR_DOC_PARSING_API_URL, PADDLEOCR_ACCESS_TOKEN and PADDLEOCR_DOC_PARSING_TIMEOUT in the environment, and python on the machine. If the API is not configured the script errors out and the skill stops — there is no fallback to the model's own vision, so check the config before pointing it at a document.
What it helps with
- pull tables out of an invoice or financial report
- read formulas from a scientific paper
- handle multi-column magazine layouts
- recover reading order on pages with diagrams
- turn scans with stamps into markdown
Example prompts for the agent
- Parse this PDF with paddleocr-doc-parsing and show me every table:
./report.pdf - Run paddleocr-doc-parsing on the document and print the full text without headers and footers
- Extract the formulas from this paper with paddleocr-doc-parsing, I need LaTeX
How to install paddleocr-doc-parsing
1Skills CLI (Vercel)
npx skills add freestylefly/canghe-skills --skill paddleocr-doc-parsing -a claude-code2Manually: put the files in place
~/.claude/skills/paddleocr-doc-parsing/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/paddleocr-doc-parsing/
1Skills CLI (Vercel)
npx skills add freestylefly/canghe-skills --skill paddleocr-doc-parsing -a codex2Manually: put the files in place
.agents/skills/paddleocr-doc-parsing/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add freestylefly/canghe-skills --skill paddleocr-doc-parsing -a cursor2Manually: put the files in place
.cursor/skills/paddleocr-doc-parsing/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install freestylefly/canghe-skills paddleocr-doc-parsing2Manually: put the files in place
.github/skills/paddleocr-doc-parsing/SKILL.md1Skills CLI (Vercel)
npx skills add freestylefly/canghe-skills --skill paddleocr-doc-parsing -a gemini-cli2Platform CLI
gemini skills install https://github.com/freestylefly/canghe-skills --path skills/paddleocr-doc-parsingManually: put the files in place
.devin/skills/paddleocr-doc-parsing/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add freestylefly/canghe-skills --skill paddleocr-doc-parsing -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about paddleocr-doc-parsing
How do I install paddleocr-doc-parsing in Claude?
Run npx skills add freestylefly/canghe-skills --skill paddleocr-doc-parsing -a claude-code. The skill ends up at ~/.claude/skills/paddleocr-doc-parsing/SKILL.md.
What does paddleocr-doc-parsing do?
It sends a document to the PaddleOCR Document Parsing API and returns its structure: text, tables, formulas, charts and reading order. The agent pulls the parts the user asked for out of the response.
What has to be set up before the first run?
The PADDLEOCR_DOC_PARSING_API_URL, PADDLEOCR_ACCESS_TOKEN and PADDLEOCR_DOC_PARSING_TIMEOUT environment variables, plus python. Without a configured API the script fails and the skill stops.
Similar skills
Everything for Office documents →paddleocr-doc-parsing by other authors comes first.
