---

name: paddleocr description: "Routes PaddleOCR users to local OCR, structured document parsing, hosted API, and training/deployment workflows." metadata: disco-role: operating disable-model-invocation: true license: Apache 2.0

PaddleOCR

PaddleOCR is a self-contained router for the repo's public local OCR, document parsing, hosted API, and maintenance workflows. Use this skill instead of reopening the source checkout for common user tasks.

Start here

• Need plain OCR, text detection/recognition, orientation, unwarping, or a single model predictor? Read sub-skills/local-ocr-pipelines/SKILL.md.
• Need PP-StructureV3, PaddleOCR-VL, PP-ChatOCRv4, PP-DocTranslation, or Office doc2md conversion? Read sub-skills/document-parsing-and-conversion/SKILL.md.
• Need hosted API calls, API auth, MCP, or LangChain integration? Read sub-skills/cloud-api-and-integrations/SKILL.md.
• Need training/config/export/deployment or TIPC evidence? Read sub-skills/training-export-and-deployment/SKILL.md.

Install and smoke checks

• Base install: python -m pip install paddleocr
• Document parsing: python -m pip install "paddleocr[doc-parser]"
• Office conversion: python -m pip install "paddleocr[doc2md]"
• Hosted API and integrations: use the package and sibling integration docs in the cloud/integrations sub-skill.
• If HuggingFace model downloads are blocked, set PADDLE_PDX_MODEL_SOURCE=BOS.

Run the bundled smoke helper after installation:

python scripts/inspect_paddleocr_env.py --cli

At minimum, this skill expects:

python -c "import paddleocr; print(paddleocr.__version__)"
paddleocr --help
paddleocr --version
paddleocr doc2md --formats

Shared references

• references/public-api-summary.md for top-level exports, CLI entry points, and option objects.
• references/installation-and-backends.md for extras, Python support, backend notes, and environment variables.
• references/troubleshooting.md for install/import, model download, auth, and output issues.
• references/repo-provenance.md for source snapshot and staleness checks.

Route boundaries

• Do not treat this root as a detailed manual. Pick the narrowest sub-skill that matches the task.
• Do not ask future agents to execute original repo scripts from the checkout. Use the bundled scripts or the sub-skill references instead.
• Do not rely on local GPU claims unless the task explicitly requires an accelerated backend and the relevant backend plan has been verified.

Refresh note

If the source repository changes, compare the new checkout against references/repo-provenance.md and refresh the sub-skill whose evidence has drifted.