【third-party】Add Claude Code Skills for PP-OCRv5 and PaddleOCR-VL by Aidenwu0209 · Pull Request #17659 · PaddlePaddle/PaddleOCR

Aidenwu0209 · 2026-02-06T05:33:43Z

概述 / Summary

添加 skills/ 目录，为 Claude Code 提供 PP-OCRv5 和 PaddleOCR-VL 的技能定义，通过百度 AI Studio API 实现 OCR 和文档解析功能。这是对现有 mcp_server/ AI 工具集成的补充。

Adds a skills/ directory providing Claude Code skill definitions for PP-OCRv5 (text extraction) and PaddleOCR-VL (document parsing) via Baidu AI Studio APIs. Complements the existing mcp_server/ AI tooling integration.

变更内容 / Changes

skills/pp-ocrv5/ - PP-OCRv5 技能：CLI 脚本、SKILL 定义、API 参考文档
skills/paddleocr-vl/ - PaddleOCR-VL 技能：CLI 脚本、SKILL 定义、API 参考文档
skills/README.md / skills/README_en.md - 中英双语文档

与 MCP Server 的关系 / Relationship to MCP Server

特性 / Feature	MCP Server	Skills
协议 / Protocol	Model Context Protocol (MCP)	Claude Code Skill Protocol
客户端 / Clients	Claude Desktop, VSCode 等	Claude Code CLI
架构 / Architecture	长驻服务进程 (stdio/HTTP)	直接 CLI 调用

两者互补：MCP Server 适用于各种 MCP 客户端，Skills 专为 Claude Code 命令行交互优化。

测试 / Testing

# 安装依赖
pip install -r skills/pp-ocrv5/scripts/requirements.txt
pip install -r skills/paddleocr-vl/scripts/requirements.txt

# 配置 API（需要 paddleocr.com 凭证）
python skills/pp-ocrv5/scripts/configure.py
python skills/paddleocr-vl/scripts/configure.py

# 运行冒烟测试
python skills/pp-ocrv5/scripts/smoke_test.py
python skills/paddleocr-vl/scripts/smoke_test.py

检查清单 / Checklist

所有 Python 文件包含 Apache 2.0 许可证头
代码通过 Black (24.10.0) 格式化
代码通过 Flake8 (7.1.1) 检查
中英双语文档（README.md / README_en.md）
无大文件（所有文件 < 512KB）
无硬编码凭证或密钥
对现有代码零影响（纯新增目录）

paddle-bot · 2026-02-06T05:33:51Z

Thanks for your contribution!

Bobholamovic

感谢贡献！留了一些建议

skills/pp-ocrv5/references/provider_api.md

skills/paddleocr-text-recognition/references/provider_api.md

skills/paddleocr-vl/references/provider_api.md

skills/paddleocr-doc-parsing/SKILL.md

skills/paddleocr-vl/SKILL.md

skills/pp-ocrv5/SKILL.md

Bobholamovic · 2026-02-06T07:56:39Z

skills/paddleocr-doc-parsing/scripts/requirements-optimize.txt

+Pillow>=10.0.0
+
+# PDF processing
+PyMuPDF>=1.23.0


建议用pypdfium2来规避license问题（pymupdf是copyleft的）

Bobholamovic · 2026-02-06T07:57:08Z

skills/paddleocr-doc-parsing/scripts/requirements.txt

@@ -0,0 +1,7 @@
+# PaddleOCR-VL 1.5 Dependencies


此处注释未对应更新。请通查整个项目，看看是否还有遗漏。

Bobholamovic · 2026-02-09T06:42:47Z

skills/paddleocr-doc-parsing/references/provider_api.md

+
+**POST** `<PADDLEOCR_VL_API_URL>`
+
+Where the URL is obtained from [Paddle AI Studio](https://paddleocr.com) (select VL model).


在我们的文档中就不要出现 Paddle AI Studio了，可以统一提 PaddleOCR official website

Bobholamovic · 2026-02-09T06:43:29Z

skills/paddleocr-doc-parsing/references/provider_api.md

+
+Where `<ACCESS_TOKEN>` is the API token obtained from Paddle AI Studio.
+
+## Request Body


这个接口描述不正确，需要参考API文档修正。parse_all等并不是API支持的参数。

Bobholamovic · 2026-02-09T06:44:47Z

skills/paddleocr-doc-parsing/references/provider_api.md

+## Best Practices
+
+1. **Use URL for large files**: Prefer `file_url` over base64 for files >5MB
+2. **Handle timeouts**: VL processing can take 3-10 seconds per page


建议只说对于大文档可能需要数分钟的处理时间

Bobholamovic · 2026-02-09T06:46:31Z

skills/paddleocr-doc-parsing/references/provider_api.md

+2. **Handle timeouts**: VL processing can take 3-10 seconds per page
+3. **Retry on 503/504**: Use exponential backoff (up to 2 retries)
+4. **Never log tokens**: Keep credentials secure
+5. **Cache responses**: Results can be cached for 10 minutes


API默认没有cache，建议去掉这一条

Bobholamovic · 2026-02-09T06:47:54Z

skills/paddleocr-doc-parsing/scripts/configure.py

+                    key = key.strip()
+                    # Skip old and new VL keys (will be overwritten)
+                    if key not in [
+                        "VL_API_URL",


这块可以不保持后向兼容，建议直接去掉对VL_API_URL和VL_TOKEN的支持，另外也去掉这个注释：“# Skip old and new VL keys (will be overwritten)”

Bobholamovic · 2026-02-09T06:48:31Z

skills/paddleocr-doc-parsing/scripts/configure.py

+                    if key not in [
+                        "VL_API_URL",
+                        "VL_TOKEN",
+                        "PADDLEOCR_VL_API_URL",


建议调整名字，不要绑定VL，而是和skill的名称对应，例如体现“文档解析”

Bobholamovic · 2026-02-09T06:48:51Z

skills/paddleocr-doc-parsing/scripts/configure.py

+                        "VL_API_URL",
+                        "VL_TOKEN",
+                        "PADDLEOCR_VL_API_URL",
+                        "PADDLEOCR_VL_ACCESS_TOKEN",


不同任务的access token通常是一样的，这里建议可以直接叫“PADDLEOCR_ACCESS_TOKEN”

Bobholamovic · 2026-02-09T06:49:53Z

skills/paddleocr-doc-parsing/scripts/configure.py

+        "VL_TOKEN", ""
+    )
+
+    print("Please provide your PaddleOCR-VL API credentials:")


需要排查整个项目中所有的文档、代码，确认PaddleOCR-VL、PP-OCRv5或类似字样不再单独出现，而是只体现“文档解析“或“文本识别”

Bobholamovic · 2026-02-09T09:25:41Z

skills/paddleocr-doc-parsing/scripts/configure.py

+            f.write("# ========================================\n")
+            f.write("# PaddleOCR Document Parsing Configuration\n")
+            f.write("# ========================================\n")
+            f.write(f"PADDLEOCR_PARSING_API_URL={api_url}\n")


要不PADDLEOCR_DOC_PARSING_API_URL吧

Bobholamovic · 2026-02-09T09:26:59Z

skills/paddleocr-doc-parsing/references/provider_api.md

+
+```json
+{
+  "file_url": "https://example.com/document.pdf"


接口还是不对，建议这块人工编写，不用AI coding工具（容易幻觉）

Bobholamovic · 2026-02-09T09:48:26Z

skills/README_en.md

+
+## Overview
+
+This directory provides two Claude Code skills for OCR text recognition and document parsing via Baidu AI Studio APIs.


建议通查所有文件，去掉“Baidu AI Studio APIs”相关表述，替换为PaddleOCR offiical API

…Metrics - Change typescript code blocks to json in output_schema.md - Remove "using PP-OCRv5" / "using PaddleOCR-VL" from directory comments - Delete unnecessary Quality Metrics section - Fix _extract_text() to handle real API response (array of pages with markdown.text) - Rewrite output_schema.md to match actual PaddleOCR-VL API response structure - Fix provider_api.md response structure documentation - Fix SKILL.md JSON examples and block labels to match real API Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Aidenwu0209 · 2026-02-10T03:19:45Z

This PR has been superseded by #17690 due to branch history issues. All review feedback has been addressed in the new PR.

paddle-bot bot added the contributor label Feb 6, 2026

Bobholamovic requested changes Feb 6, 2026

View reviewed changes

Aidenwu0209 force-pushed the add-claude-code-skills branch 4 times, most recently from 92fafc5 to 084c906 Compare February 6, 2026 10:55

Bobholamovic requested changes Feb 9, 2026

View reviewed changes

Aidenwu0209 force-pushed the add-claude-code-skills branch from 084c906 to a6fd5d2 Compare February 9, 2026 08:48

Bobholamovic requested changes Feb 9, 2026

View reviewed changes

Aidenwu0209 force-pushed the add-claude-code-skills branch 2 times, most recently from 6cd19ac to 3e18a70 Compare February 9, 2026 10:32

Aidenwu0209 closed this Feb 9, 2026

Aidenwu0209 force-pushed the add-claude-code-skills branch from 3e18a70 to f773f2c Compare February 9, 2026 14:56


		POST `<PADDLEOCR_VL_API_URL>`

		Where the URL is obtained from [Paddle AI Studio](https://paddleocr.com) (select VL model).


		Where `<ACCESS_TOKEN>` is the API token obtained from Paddle AI Studio.

		## Request Body


		## Overview

		This directory provides two Claude Code skills for OCR text recognition and document parsing via Baidu AI Studio APIs.

Conversation

Aidenwu0209 commented Feb 6, 2026

概述 / Summary

变更内容 / Changes

与 MCP Server 的关系 / Relationship to MCP Server

测试 / Testing

检查清单 / Checklist

Uh oh!

paddle-bot bot commented Feb 6, 2026

Uh oh!

Bobholamovic left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Aidenwu0209 commented Feb 10, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants