- Add 7 new Word tools to README (outline, search, entities, etc.)
- Add 9 MCP prompts section with workflow descriptions
- Gitignore reading progress bookmark files (.*.reading_progress.json)
- Gitignore local .mcp.json and test documents
New tools for Word document analysis:
- extract_entities: Pattern-based extraction of people, places, organizations
- get_chapter_summaries: Chapter previews with opening sentences and word counts
- save_reading_progress: Bookmark reading position to JSON file
- get_reading_progress: Resume reading from saved position
New MCP prompts (basic to advanced workflows):
- explore-document: Get started with a new document
- find-character: Track character mentions
- chapter-preview: Quick chapter overviews
- resume-reading: Continue where you left off
- document-analysis: Comprehensive multi-tool analysis
- character-journey: Track character arc through narrative
- document-comparison: Compare entities between chapters
- full-reading-session: Guided reading with bookmarking
- manuscript-review: Complete editorial workflow
Updated test counts for 19 total tools (6 universal + 10 word + 3 excel)
New tools for easier document navigation:
- get_document_outline: Structured view of headings with chapter detection
- check_style_consistency: Find formatting issues and missing chapters
- search_document: Search with context and chapter location
All tools tested with 200+ page manuscript. Detects issues like
Chapter 3 being styled as "normal" instead of "Heading 1".
Documents 6 critical bugs discovered while processing a 200+ page
manuscript, including the root cause xpath API mismatch between
python-docx and lxml that caused silent failures in chapter search.
python-docx elements don't support xpath() with namespaces kwarg.
The calls silently failed in try/except blocks, causing chapter search
and heading detection to never find matches.
Fixed by replacing xpath(..., namespaces={...}) with:
- findall('.//' + qn('w:t')) for text elements
- find(qn('w:pPr')) + find(qn('w:pStyle')) for style detection
- get(qn('w:val')) for attribute values
Also fixed logic bug where elif prevented short-text fallback from
running when a non-heading style existed on the paragraph.
The pyproject.toml script entry point (mcp-office-tools) uses server.py,
not server_monolithic.py. Applied same show_banner=False fix and
simplified to use app.run() instead of asyncio.run(app.run_stdio_async()).
Bug fixes:
- Remove 100-paragraph cap that prevented extracting content past ~page 4
Now calculates limit based on number of pages requested (300 paras/page)
- Add fallback page estimation when docs lack explicit page breaks
Uses ~25 paragraphs per page for navigation in non-paginated docs
- Fix _get_available_headings to scan full document (was only first 100 elements)
Headings like Chapter 10 at element 1524 were invisible
- Fix MCP connection by disabling FastMCP banner (show_banner=False)
ASCII art banner was corrupting stdout JSON-RPC protocol
Changes:
- Default image_mode changed from 'base64' to 'files' to avoid huge responses
- Add proper .mcp.json config with command/args format
- Add test document to .gitignore for privacy
Links README to Ryan's AI discernment article, which discusses
the documentation rewrite process and connects to the model's
perspective in the collaborations archive.
- Replace generic opener with direct description
- Make feature bullets more conversational (less "feature list" mode)
- Add context before format support table
- Clarify pagination example with "three ways" structure
- Lead testing section with the dashboard hook
- Add architecture design rationale
- Remove "comprehensive" and "intelligent" buzzwords
- Document all 12 actual MCP tools (6 universal, 3 Word, 3 Excel)
- Add comprehensive format support matrix with feature breakdown
- Include practical usage examples with real output structures
- Add test dashboard section
- Simplify installation with uvx/Claude Code instructions
- Remove marketing fluff; focus on technical accuracy
- Self-contained HTML dashboard with MS Office 365 design
- pytest plugin captures inputs, outputs, and errors per test
- Unified orchestrator runs pytest + torture tests together
- Test files persisted in reports/test_files/ with relative links
- GitHub Actions workflow with PR comments and job summaries
- Makefile with convenient commands (test, view-dashboard, etc.)
- Works offline with embedded JSON data (no CORS issues)
- Create @resolve_field_defaults decorator to handle Pydantic FieldInfo
objects when tools are called directly (outside MCP framework)
- Create @handle_office_errors decorator for consistent error wrapping
- Apply decorators to Excel and Word mixins, removing ~100 lines of
boilerplate code
- Fix Excel formula extraction performance: load workbooks once before
loop instead of per-cell (100x faster with calculated values)
- Update test suite to use correct mock patch paths (patch where names
are looked up, not where defined)
- Add torture_test.py for real document validation
- Use app.run_stdio_async() instead of deprecated stdio_server import
- Aligns with FastMCP 2.11.3 API
- Server now starts correctly with uv run mcp-office-tools
- Maintains all MCPMixin functionality and tool registration
- Add main() function back to server.py for CLI script entry point
- Maintains FastMCP MCPMixin pattern while fixing uvx execution
- Server now starts properly with 'uvx --from . mcp-office-tools'
- Preserves all 7 tools with official mixin registration
- Split monolithic 2209-line server.py into organized mixin classes
- UniversalMixin: Format-agnostic tools (extract_text, extract_images, etc.)
- WordMixin: Word-specific tools (convert_to_markdown with chapter_name support)
- ExcelMixin: Placeholder for future Excel-specific tools
- PowerPointMixin: Placeholder for future PowerPoint-specific tools
Benefits:
• Improved maintainability and separation of concerns
• Better testability with isolated mixins
• Easier team collaboration on different file types
• Reduced cognitive load per module
• Preserved all 7 existing tools with full functionality
Architecture now supports clean expansion for format-specific tools
while maintaining backward compatibility through legacy server backup.
- Add chapter_name parameter to convert_to_markdown tool
- Implement _find_chapter_content_range() for heading-based navigation
- Add _get_available_headings() to help users find chapter names
- Include chapter extraction metadata in results
- Enhanced ultra-fast summary with available headings
- Provides alternative to bookmark extraction when bookmarks unavailable
- Add bookmark_name parameter for extracting specific chapters/sections
- Implement bookmark boundary detection using Word XML structure
- Extract content between bookmark start/end markers with smart extension
- More reliable than page ranges - bookmarks are anchored to exact locations
- Support chapter extraction like bookmark_name='Chapter1_Start'
- Include bookmark metadata in response with element ranges
- Perfect for extracting individual chapters from large documents
- Replace unreliable Word page detection with element-based limiting
- Cap extraction at 25 paragraphs per 'page' requested (max 100 total)
- Cap extraction at 8k chars per 'page' requested (max 40k total)
- Add early termination when limits reached
- Add processing_limits metadata to show actual extraction stats
- Prevent 1.28M token responses by stopping at reasonable content limits
- Single page (page_range='1') now limited to ~25 paragraphs/8k chars
- Bypass all complex processing in summary_only mode
- Extract only first 50 paragraphs, max 10 headings, 5 content paragraphs
- Add bookmark detection for chapter navigation hints
- Limit summary content to 2000 chars max
- Prevent 1,282,370 token responses with surgical precision
- Show bookmark names as chapter start indicators
- Extract headings with page numbers during document processing
- Generate optimized page ranges for each section/chapter
- Provide intelligent chunking suggestions (15-page optimal chunks)
- Classify section types (chapter, section, subsection, etc.)
- Calculate actual section lengths based on heading positions
- Include suggested_chunking with ready-to-use page ranges
- Perfect for extracting 200+ page documents section by section
- Analyze document size and complexity before processing
- Provide clear workflow recommendations in response metadata
- Strongly recommend summary_only + page_range for large documents (>10 pages)
- Add warning system for suboptimal usage patterns
- Update parameter descriptions with best practice guidance
- Help users avoid 25k token response limits proactively
- Add page break detection using Word XML structure
- Process only specified pages instead of full document + truncation
- Route page-range requests to python-docx for granular control
- Skip mammoth for page-specific processing (mammoth processes full doc)
- Add page metadata to results when filtering is used
- Significantly reduce memory usage and response size for large documents
- Add eye-catching visual design with emojis and badges
- Create compelling hero section with value proposition
- Include real-world benchmarks and performance metrics
- Add enterprise success stories and use cases
- Implement collapsible sections for better organization
- Include Mermaid architecture diagram
- Add comprehensive feature matrix with visual indicators
- Create roadmap and community sections
- Enhance installation and setup instructions
- Make it GitHub-ready with proper formatting
🚀 Now ready to wow potential users and contributors!
- Comprehensive Microsoft Office document processing server
- Support for Word (.docx, .doc), Excel (.xlsx, .xls), PowerPoint (.pptx, .ppt), CSV
- 6 universal tools: extract_text, extract_images, extract_metadata, detect_office_format, analyze_document_health, get_supported_formats
- Multi-library fallback system for robust processing
- URL support with intelligent caching
- Legacy Office format support (97-2003)
- FastMCP integration with async architecture
- Production ready with comprehensive documentation
🤖 Generated with Claude Code (claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
