asset-aware-mcp

🏥 Medical RAG with Asset-Aware MCP - Precise PDF asset retrieval (tables, figures, sections) and Knowledge Graph for AI Agents.

🌐 繁體中文

🎯 Why Asset-Aware MCP?

AI cannot directly read image files on your computer. This is a common misconception.

Real-world Effect

# After retrieving the image via MCP, the AI can analyze it directly:

User: What is this figure about?

AI: This is the architecture diagram for Scaled Dot-Product Attention:
    1. Inputs: Q (Query), K (Key), V (Value)
    2. MatMul of Q and K
    3. Scale (1/√dₖ)
    4. Optional Mask (for decoder)
    5. SoftMax normalization
    6. Final MatMul with V to get the output

This is the value of Asset-Aware MCP - enabling AI Agents to truly "see" and understand charts and tables in your PDF literature.

✨ Features

• 📄 Asset-Aware ETL - PDF → Markdown with dual-engine PDF parsing:
  • PyMuPDF (default) - Fast extraction (~50MB)
  • Marker (optional, use_marker=True) - High-precision structured parsing with blocks.json (bbox/coordinates)
• 🧩 Unified Segmentation Export - Normalized segmentation.json merges manifest, blocks, reading order, and persisted markdown line spans for downstream tools and extensions.
• 🖼️ Layout Overlay Debugging - Render page overlays from original.pdf to inspect bbox, segment type, and reading order visually.
• 🔤 On-Demand OCR Preprocessing - Optional ocrmypdf preprocessing path for scanned PDFs before ETL.
• 🧭 Section Navigation - Dynamic hierarchy section tree with 5 tools: browse, search, detail, content reading, and block extraction for any depth of headings.
• 🔄 Async Job Pipeline - Supports asynchronous task processing and progress tracking for large documents.
• 🗺️ Document Manifest - Provides a structured "map" of the document for precise data access by Agents.
• 🧠 LightRAG Integration - Knowledge Graph + Vector Index, supporting cross-document comparison and reasoning.
• 🧾 Citation-Aware KG Output - consult_knowledge_graph now supports structured answer/reference payloads for downstream agent workflows.
• 📝 Docx Editing (DFM) - Edit .docx files in Markdown via Docx-Flavored Markdown format. Supports legacy .doc, .odt, and .ods ingest via LibreOffice auto-conversion. 16 tools: ingest, read, save, list, delete, export, strict round-trip validation, DOCX→PDF, DOCX→DOC, DOCX→ODT, and Docx ↔ A2T bridges.
• 🛡️ DFM Integrity Checker - Automatic validation and auto-repair at every pipeline stage (post-ingest, pre-save, post-save). Catches orphan markers, column mismatches, and format inconsistencies.
• 📊 A2T (Anything to Table) - 7 operation-based tools for building professional tables from any source (PDF assets, Knowledge Graph, URLs, user input). Features: Citations (AssetRef), Audit Trail, Schema Evolution, Templates, Drafting, and Token-efficient resumption.
• 🖥️ VS Code Management Extension - Graphical interface for monitoring server status, ingested documents, and A2T tables/drafts with one-click Excel export.
• 🔌 MCP Server - Exposes tools and resources to Copilot/Claude via FastMCP.
• 🏥 Medical Research Focus - Optimized for medical literature, supporting Base64 image transmission for Vision AI analysis.

🏗️ Architecture

┌─────────────────────────────────────────────────────────┐
│                    AI Agent (Copilot)                   │
└─────────────────────┬───────────────────────────────────┘
                      │ MCP Protocol (Tools & Resources)
┌─────────────────────▼───────────────────────────────────┐
│            MCP Server (Modular Presentation)            │
│  ┌─────────────────────────────────────────────────┐   │
│  │ tools/: 59 tools in 7 modules                   │   │
│  │   document (18) │ docx (16) │ section (5)       │   │
│  │   job (4) │ knowledge (3) │ table (7) │ profile (6) │
│  └─────────────────────────────────────────────────┘   │
│  ┌─────────────────────────────────────────────────┐   │
│  │ resources/: 13 resources in 2 modules           │   │
│  └─────────────────────────────────────────────────┘   │
└─────────────────────┬───────────────────────────────────┘
                      │
┌─────────────────────▼───────────────────────────────────┐
│                  ETL Pipeline (DDD)                     │
│  ┌──────────┐  ┌──────────┐  ┌──────────┐              │
│  │ PyMuPDF  │  │  Asset   │  │ LightRAG │              │
│  │ Adapter  │→ │  Parser  │→ │  Index   │              │
│  └──────────┘  └──────────┘  └──────────┘              │
└─────────────────────┬───────────────────────────────────┘
                      │
┌─────────────────────▼───────────────────────────────────┐
│                   Local Storage                         │
│  ./data/                                                │
│  ├── doc_{id}/        # Document Assets                 │
│  ├── docx_{id}/       # Docx IR + DFM + Assets          │
│  ├── tables/          # A2T Tables (JSON/MD/XLSX)       │
│  │   └── drafts/      # Table Drafts (Persistence)      │
│  └── lightrag_db/     # Knowledge Graph                 │
└─────────────────────────────────────────────────────────┘

📁 Project Structure (DDD)

asset-aware-mcp/
├── src/
│   ├── domain/              # 🔵 Domain: Entities, Value Objects, Interfaces
│   ├── application/         # 🟢 Application: Doc Service, Table Service (A2T), Asset Service
│   ├── infrastructure/      # 🟠 Infrastructure: PyMuPDF, LightRAG, Excel Renderer
│   └── presentation/        # 🔴 Presentation: MCP Server (FastMCP)
├── data/                    # Document and Asset Storage
├── docs/
│   └── spec.md              # Technical Specification
├── tests/                   # Unit and Integration Tests
├── vscode-extension/        # VS Code Management Extension
└── pyproject.toml           # uv Project Config

📐 Architecture Diagrams

Visual overview for the project. All diagrams use consistent GitHub README style.

💡 All generation prompts are saved in docs/diagrams/ALL-PROMPTS.md for style consistency and regeneration.

🚀 Quick Start

# Install dependencies (using uv) — default install skips Marker/torch
uv sync

# v0.6.27: Marker extra is temporarily empty because marker-pdf pins
# Pillow<11 while the secure runtime requires Pillow>=12.2.0.
# Use the default PyMuPDF backend until upstream marker-pdf supports patched Pillow.

# Run MCP Server
uv run python -m src.presentation.server

# Or use the VS Code extension for graphical management

Runtime note: The VS Code extension prefers a managed Python 3.11 runtime when launching the MCP server via uv or uvx. This avoids native package builds on end-user machines, especially macOS systems without Xcode Command Line Tools, while keeping the project itself compatible with newer Python versions.

Installation scope note:

• The VS Code extension installs once per user (global). The MCP server launched through uvx asset-aware-mcp reuses the user uv cache rather than reinstalling per workspace.
• Runtime data stays with your repo: .env and assetAwareMcp.dataDir default to ./data, so ingested assets remain scoped to the current workspace.

Marker note: In v0.6.27 the packaged Marker extra is intentionally on security hold: upstream marker-pdf 1.10.2 requires Pillow<11, while this release pins Pillow>=12.2.0 for patched image-processing security. Default installs use the PyMuPDF backend only. use_marker=True / parse_pdf_structure will report that Marker is unavailable until upstream Marker supports a patched Pillow range.

🔌 MCP Tools

Document & Asset Tools

Job Management Tools

Knowledge Graph Tools

Knowledge graph note:

• consult_knowledge_graph defaults to response_mode="structured" and can return answer, references, metadata, retrieval, and counts for agent-side citation workflows.
• Use response_mode="data" when you want retrieval payloads without final answer synthesis, or response_mode="text" for legacy plain-text behavior.

Section Navigation Tools (Dynamic Hierarchy)

Docx Editing Tools (DFM — Docx-Flavored Markdown)

Edit .docx files as Markdown. Preserves formatting, tables, media on round-trip.

A2T (Anything to Table) Tools — 7 Operation-Based Tools

Agent-friendly design: each tool handles multiple operations via operation parameter. Tables accept any source — PDF assets, KG entities, external URLs, or user input.

ETL Profile Tools

Different journals/formats need different extraction settings. Use these tools to switch profiles.

🔧 Tech Stack

📋 Documentation

Installation guidance:

• Default install: uv sync

• Marker backend: temporarily disabled in v0.6.27 because marker-pdf pins vulnerable Pillow<11; the marker / pdf extras are compatibility placeholders until upstream supports patched Pillow.

• VS Code extension: assetAwareMcp.enableMarkerBackend is retained as a setting, but the launcher will not install marker-pdf while the security hold is active.

• Technical Spec - Detailed technical specification

• Architecture - System architecture

• Constitution - Project principles

• Competitive Analysis - MCP + DOCX ecosystem landscape

📄 License

Apache License 2.0