mirror of
https://github.com/PaddlePaddle/PaddleOCR.git
synced 2026-08-29 04:12:51 +08:00
Lin Manhui
1e5aa0ad31
* refactor(mcp): add Executor base class and module structure Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * fix(code quality): rename TimeoutError and remove unnecessary pass statements - Rename TimeoutError to ExecutionTimeoutError to avoid shadowing built-in exception - Remove unnecessary pass statements from exception classes and abstract methods Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * refactor(mcp): add HTTPExecutor base class for synchronous HTTP APIs * feat(mcp): add AIStudioExecutor using AsyncPaddleOCRClient Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * refactor(mcp): add LocalExecutor for local inference - Add LocalExecutor class for local PaddleOCR inference - Add _EngineWrapper to run synchronous engine in async context - Support all pipeline types: OCR, PP-StructureV3, PaddleOCR-VL series - Handle file path, URL, and Base64 input formats Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * refactor(mcp): add QianfanExecutor and SelfHostedExecutor - QianfanExecutor: Bearer token authentication, "paddleocr" endpoint - SelfHostedExecutor: No authentication, "ocr"/"layout-parsing" endpoints - Both inherit HTTPExecutor with unified response parsing Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * refactor(mcp): add executor factory function Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * refactor(mcp): add capabilities module with OCR and layout parsing Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * refactor(mcp): fix type hints and remove pass statements in capabilities * refactor(mcp): update __main__.py to use new architecture * refactor(mcp): update copyright year to 2026 * refactor(mcp): add type hint to main() function * chore(mcp): bump version to 0.7.0 * refactor(mcp): remove old pipelines.py * docs(mcp): update docs for v0.7.0 SDK-based architecture - Remove PADDLEOCR_MCP_SERVER_URL for aistudio mode (SDK has built-in default) - Add optional PADDLEOCR_MCP_BASE_URL for custom AI Studio base URL - Update parameter reference table * Update code * fix(mcp): fix AIStudioExecutor SDK integration - Use PPStructureV3Options for PP-StructureV3 pipeline - Use PaddleOCRVLOptions for PaddleOCR-VL series - Fix DocParsingPage attribute names (markdown_text, markdown_images) * refactor(mcp): rename layout to doc_parsing - Rename layout.py to doc_parsing.py - Rename LayoutParsingCapability to DocParsingCapability - Rename _parse_layout_response_http to _parse_doc_parsing_response_http - Rename _parse_layout_result to _parse_doc_parsing_result - Remove file path comments from all Python files * docs(mcp): make DocParsingCapability docstring more generic - Change from listing specific pipelines to "e.g." style - Make tool_name description more generic for future extensibility * remove unused files * docs(mcp): update terminology and remove obsolete URL configuration - Change '工作模式' to '推理方式' / 'Inference methods' - Unify with SDK docs: use '官方 API' / 'Official API' - Update options: 本地推理、官方API、千帆API、自建API - Remove obsolete manual URL configuration (SDK has built-in defaults) - Update token acquisition instructions * docs(mcp): unify English doc terminology - working modes/mode → inference methods/method/API - Update TOC: 'Working Modes Explained' → 'Inference Methods' - Change 'Mode 1: Local Python Library' → 'Method 1: Local Inference' - Update parameter table: 'mode' → 'source', 'service' → 'API' - 'local Python library' → 'local inference' - 'PaddleOCR official service' → 'Official API' - 'Qianfan platform service' → 'Qianfan API' - 'self-hosted service' → 'Self-hosted API' - Update example command comments - Update 'Local mode' → 'Local Inference' in uvx section - Update Known Limitations section * fix(mcp): correct inaccuracies and unify terminology in docs and code - Remove incorrect Qianfan pipeline validation (OCR is supported) - Fix broken anchor link in Chinese doc 2.4 section - Unify Chinese doc terminology: - 千帆平台服务/自托管服务 → 千帆 API/自建 API - PaddleOCR 官方服务 → 官方 API - 本地 Python 库模式 → 本地推理 - 模式/方式 → 推理方式 - Fix Chinese doc numbered list error (5 → 4) - Remove incorrect Qianfan pipeline limitation note from English doc - Update example command comments to use consistent terminology * docs(mcp): simplify pipeline documentation and remove hardcoded lists - Replace Chinese docstrings with English in base.py - Remove hardcoded pipeline enumeration from AIStudioExecutor docstring - Simplify pipeline table: merge PaddleOCR-VL series into single entry - Add 'etc.' to parameter options to indicate extensibility * Fix bugs * refactor(mcp): let SDK use service-side defaults - Remove hardcoded parameter defaults (use_doc_orientation_classify=False, etc.) - Only set visualize=False (MCP tool never returns visualizations) - Pass-through all optional parameters from user options dict - This allows server-side optimizations and better future compatibility * feat(mcp): add reasonable default options - All pipelines: use_doc_orientation_classify=False, use_doc_unwarping=False - PP-StructureV3: use_chart_recognition=True - PaddleOCR-VL series: use_chart_recognition=True, use_seal_recognition=True - Users can override these defaults via options dict * refactor(mcp): move default options to MCPCapability - Move default options from AIStudioExecutor to MCPCapability._process - All executors now receive same default behavior via capability layer - User options override defaults - Add pipeline parameter to MCPCapability and subclasses * refactor(mcp): redesign capabilities as pipeline-level classes - MCPCapability: add PIPELINE class attribute, remove constructor param - BaseDocParsingCapability: new base class for doc parsing pipelines - PPStructureV3Capability: dedicated class for PP-StructureV3 - PaddleOCRVLCapability: handles PaddleOCR-VL series with pipeline param - OCRCapability: define PIPELINE as class attribute - Remove old DocParsingCapability (replaced by specific classes) - Factory simplified: no need to pass pipeline to constructors * refactor(mcp): use method for default options instead of hardcoded logic - MCPCapability: add DEFAULT_OPTIONS constant and get_default_options method - BaseDocParsingCapability: override get_default_options for chart recognition - PaddleOCRVLCapability: override get_default_options for seal recognition - Move default logic from hardcoded if/else to method inheritance - PaddleOCRVLCapability: set PIPELINE dynamically in __init__ * Big refactor * Fix bugs * Remove tests * Remove unused comments * Fix doc bugs --------- Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>