simple-knowledge-graph-mcp

Eleftheria14/simple-knowledge-graph-mcp

3.1

If you are the rightful owner of simple-knowledge-graph-mcp and would like to certify it and/or have it hosted online, please leave a comment on the right or send an email to dayong@mcphub.com.

This is a Model Context Protocol (MCP) server that transforms documents into an intelligent, searchable knowledge base.

Tools
5
Resources
0
Prompts
0

GROBID-Powered Academic Knowledge Graph MCP

Transform research papers into intelligent, queryable knowledge graphs using Claude Desktop with zero API costs and superior academic understanding.

What This Does

This is a GROBID-powered Model Context Protocol (MCP) server specifically designed for academic research. It transforms research papers into an intelligent, searchable knowledge base with superior citation network analysis. It provides Claude Desktop with 5 specialized tools to extract, store, and query academic information using advanced academic document processing.

Core Architecture

  • GROBID Academic Parser: Self-hosted service providing 87-90% F1-score accuracy for research papers
  • Neo4j Unified Database: Stores entities, relationships, and vector embeddings in one database
  • FastMCP Server: Orchestrates 5 academic-focused tools for Claude Desktop
  • Zero API Costs: Self-hosted GROBID eliminates expensive PDF processing fees
  • Academic Focus: Built specifically for research papers, not generic documents

Academic Processing Architecture

flowchart TB
    subgraph clients["🖥️ LLM Clients"]
        A[Claude Desktop]
        B[Other MCP Clients]
    end
    
    subgraph services["🐳 Academic Services"]
        GROBID[GROBID Docker<br/>Academic PDF Parser<br/>87-90% F1-Score]
    end
    
    subgraph mcp["📡 MCP Server"]
        E[FastMCP Server<br/>Academic Tools]
    end
    
    subgraph academic_flow["📚 ACADEMIC PROCESSING WORKFLOW"]
        direction TB
        F[extract_and_store_entities<br/>Academic Entity Extraction]
        G[store_vectors<br/>Neo4j Vector Storage]
        H[grobid_extract<br/>Academic PDF Processing]
        
        subgraph grobid_pipeline["GROBID Pipeline"]
            H1[Authors & Affiliations] 
            H2[Citations & References]
            H3[Academic Structure]
            H4[Tables & Figures]
        end
        
        subgraph entity_pipeline["Academic Entity Pipeline"]
            F1[Research Concepts] --> F2[Author Networks]
            F2 --> F3[Citation Relationships]
            F3 --> F4[Neo4j Graph Store]
        end
        
        subgraph vector_pipeline["Vector Pipeline"] 
            G1[Academic Text Chunks] --> G2[Local Embeddings] --> G3[Neo4j Vector Store]
        end
        
        H --> H1
        H --> H2  
        H --> H3
        H --> H4
        F --> F1
        G --> G1
    end
    
    subgraph database["🗄️ Unified Knowledge Store"]
        NEO@{shape: database, label: "Neo4j Database<br/>• Academic Entities<br/>• Citation Networks<br/>• Vector Embeddings<br/>• Research Relationships"}
    end
    
    subgraph query_flow["🔍 ACADEMIC QUERYING WORKFLOW"]
        direction TB
        I[query_knowledge_graph<br/>Academic Search]
        J[generate_literature_review<br/>Citation Formatting]
        
        subgraph academic_rag["Academic GraphRAG"]
            I1[Author Network<br/>Traversal] 
            I2[Citation Network<br/>Analysis]
            I3[Semantic Research<br/>Search]
            I4[Combine Academic<br/>Results]
        end
        
        I --> I1
        I --> I2
        I --> I3
        I1 --> I4
        I2 --> I4
        I3 --> I4
        I4 --> J
    end
    
    %% Client connections
    A --> E
    B --> E
    
    %% Service connections
    GROBID --> H
    
    %% Processing flow
    E --> H
    E --> F
    E --> G
    F4 --> NEO
    G3 --> NEO
    
    %% Query flow
    E --> I
    E --> J
    NEO --> I1
    NEO --> I2
    NEO --> I3
    
    %% Styling
    classDef academicBox fill:#e8f5e8
    classDef serviceBox fill:#fff3e0
    classDef queryBox fill:#f3e5f5
    classDef dbBox fill:#e1f5fe
    
    class academic_flow,grobid_pipeline,entity_pipeline,vector_pipeline academicBox
    class services,GROBID serviceBox
    class query_flow,academic_rag queryBox
    class database,NEO dbBox

How Academic Data Flows Through the System

📚 Academic Processing Workflow:

  1. GROBID extracts structured data → grobid_extract → Authors, citations, academic structure
  2. LLM analyzes academic content → extract_and_store_entities → Research concepts, author networks
  3. Academic text storage → store_vectors → Local embeddings → Neo4j Vector Store

🔍 Academic Query Workflow (Academic GraphRAG):

  1. Research question → query_knowledge_graph
  2. Multi-modal academic search: Author networks + Citation analysis + Semantic research search
  3. Academic result synthesis → Comprehensive answer with proper citations
  4. Literature review generation → generate_literature_review → Formatted academic output

🗄️ Unified Academic Knowledge Store:

  • Neo4j Database: Academic entities, citation networks, author relationships, and vector embeddings all in one system

Step-by-Step: How Academic Processing Works

1. Research Paper Upload & Academic Processing

You → Upload PDFs to Claude Desktop
Claude → Uses GROBID for superior academic extraction

What GROBID + Claude extracts:

  • Academic Entities: Authors, institutions, research concepts, methodologies with confidence scores
  • Citation Networks: Paper-to-paper references, author collaboration patterns
  • Academic Structure: Abstracts, sections, tables, figures with proper academic context
  • Research Relationships: How concepts, authors, and papers interconnect across the literature

2. Academic Data Storage

Academic Entities + Citation Networks → Neo4j Graph Database
Research Text + Academic Context → Neo4j Vector Store with local embeddings

Example academic data stored:

// Academic Entity (Author)
{
  "id": "hinton_2006",
  "name": "Geoffrey Hinton", 
  "type": "person",
  "properties": {
    "affiliation": "University of Toronto",
    "research_areas": ["deep learning", "neural networks"],
    "h_index": 180
  },
  "confidence": 0.95
}

// Citation Relationship  
{
  "source": "hinton_2006_paper",
  "target": "lecun_1998_paper",
  "type": "cites",
  "context": "Building on LeCun's convolutional architecture...",
  "citation_count": 1
}

// Academic Text Vector (in Neo4j)
{
  "text": "Our proposed attention mechanism allows the model to focus...",
  "embedding": [0.1, -0.3, 0.8, ...],
  "document_title": "Attention Is All You Need",
  "section": "methodology",
  "authors": ["Vaswani, A.", "Shazeer, N."],
  "year": 2017
}

3. Academic Intelligence Querying

You → Ask research questions in natural language
Claude → Uses academic MCP tools to search citation networks + semantic content
Claude → Returns comprehensive academic answers with proper citations

Academic query types supported:

  • Author network analysis: "Who are the key collaborators in transformer research?"
  • Citation relationship mapping: "How do attention mechanisms relate to BERT?"
  • Research concept evolution: "How has the transformer architecture evolved since 2017?"
  • Cross-paper methodology analysis: "What are the different approaches to fine-tuning in these papers?"
  • Academic literature reviews: "Generate a comprehensive review of attention mechanisms with citations"

4. Academic Knowledge Graph Evolution

More research papers → Richer citation networks → Deeper academic insights

As you add research papers:

  • Citation networks emerge: GROBID identifies when papers reference each other with high accuracy
  • Author collaboration networks form: Maps institutional and research partnerships
  • Research concept evolution tracked: Shows how methodologies develop across publications
  • Research gaps identified: Highlights understudied areas and potential research opportunities

The 5 Academic MCP Tools Available to Claude

  1. extract_and_store_entities - Extract academic entities (authors, concepts, methodologies) and relationships from research content
  2. store_vectors - Store research text chunks as vectors in Neo4j with academic context preservation
  3. query_knowledge_graph - Search both citation networks and semantic content for comprehensive academic answers
  4. generate_literature_review - Format academic results with proper citations (APA, IEEE, Nature, MLA)
  5. clear_knowledge_graph - Reset all academic data for fresh research projects

Real-World Academic Workflow

Scenario: Building a Transformer Research Knowledge Base

Step 1: Research Paper Collection

# Upload 20 transformer research papers to Claude Desktop project
- attention_is_all_you_need.pdf (Vaswani et al., 2017)
- bert_pretraining.pdf (Devlin et al., 2018)
- gpt_language_models.pdf (Radford et al., 2018)
- ... (17 more seminal transformer papers)

Step 2: GROBID-Powered Academic Extraction

You: "Process these transformer papers with academic extraction"

Claude: *Uses GROBID + extract_and_store_entities*
- GROBID extracts 150+ authors with institutional affiliations
- Maps 300+ research concepts (attention, positional encoding, etc.)
- Identifies 200+ citation relationships between papers
- Stores everything in Neo4j with 87-90% accuracy

You: "Store the key research passages with academic context"

Claude: *Uses store_vectors tool*  
- Stores 2000+ academic text chunks with section context
- Preserves author information and publication details
- Enables semantic search across methodologies and results

Step 3: Academic Intelligence Analysis

You: "What are the key innovations in transformer architectures across these papers?"

Claude: *Uses query_knowledge_graph tool*
- Searches Neo4j citation networks for transformer innovations
- Finds semantic matches in academic text vectors
- Returns: Self-attention mechanisms, positional encoding, multi-head attention, layer normalization
- Includes proper citations with institutional context (Google Research, OpenAI, etc.)

You: "Generate a literature review on attention mechanisms evolution"

Claude: *Uses generate_literature_review tool*
- Organizes findings by chronological development
- Formats with proper academic citations (APA style)
- Includes author collaboration patterns and research trends
- Shows institutional contributions to attention mechanism development

Step 4: Advanced Academic Queries

You: "Which researchers have collaborated most frequently in transformer research?"

Claude: Maps collaboration networks from GROBID-extracted author relationships
Result: Shows Vaswani-Shazeer collaborations, Google Research teams, etc.

You: "What are the different approaches to positional encoding across these papers?"

Claude: Finds methodological variations using semantic search + citation analysis
Result: Sinusoidal encoding (Vaswani), learned embeddings (BERT), relative positioning

You: "Trace the evolution of attention mechanisms from 2017-2023"

Claude: Temporal analysis through citation networks and concept development
Result: Self-attention → Multi-head → Sparse attention → Efficient transformers

Key Academic Benefits

For Researchers

  • Superior PDF processing: GROBID provides 87-90% F1-score accuracy vs generic tools (60-70%)
  • Zero API costs: Self-hosted GROBID eliminates expensive LlamaParse fees ($10-30/month)
  • Citation network analysis: Map author collaborations and institutional partnerships automatically
  • Research gap identification: Discover understudied areas through comprehensive literature mapping
  • Academic literature reviews: Generate properly formatted reviews with correct citations

For Graduate Students

  • Thesis research acceleration: Build comprehensive knowledge bases from research literature
  • Academic writing support: Automatic citation extraction with proper formatting (APA, IEEE, Nature, MLA)
  • Methodology comparison: Understand how research approaches connect across different papers
  • Advisor collaboration: Share queryable knowledge graphs with supervisors

For Academic Institutions

  • Research intelligence: Track institutional research output and collaboration patterns
  • Literature monitoring: Stay current with developments in specific research areas
  • Grant application support: Identify research trends and potential collaborators
  • Publication analysis: Map citation networks and research impact across departments

Technical Architecture

Academic Processing Layer (src/processor/)

processor/
├── document_pipeline.py        # LangGraph orchestrator for academic workflows
├── tools/
│   ├── grobid_tool.py          # GROBID academic PDF extraction (87-90% accuracy)
│   └── storage_tool.py         # Neo4j integration for academic data
├── config.py                   # Processing configuration management
├── orchestrator_config.py      # Academic workflow definitions
└── entity_extractor_config.py  # Academic extraction settings

Storage Layer (src/storage/)

storage/
├── neo4j/           # Unified graph and vector database
│   ├── storage.py   # Academic entities, citations, and vector storage
│   └── query.py     # Citation network traversal and semantic search
└── embedding/       # Local embedding generation
    └── service.py   # sentence-transformers integration (privacy-focused)

Tools Layer (src/tools/)

tools/
├── storage/         # Academic data persistence tools
│   ├── enhanced_entity_storage.py   # Academic entity extraction MCP tool
│   ├── neo4j_vector_storage.py      # Vector storage in Neo4j MCP tool
│   └── database_management.py       # clear_knowledge_graph MCP tool
├── query/           # Academic data retrieval tools
│   ├── knowledge_search.py          # query_knowledge_graph MCP tool (citation networks + semantic)
│   └── literature_generation.py     # generate_literature_review MCP tool
└── shared_registry.py               # Dual MCP/LangChain tool registry

Server Layer (src/server/)

  • main.py - FastMCP server with all 5 academic tools registered for Claude Desktop

Academic-First Privacy Design

  • Self-hosted GROBID - No external APIs for PDF processing, complete privacy
  • Local embeddings - Uses sentence-transformers for complete research confidentiality
  • Neo4j via Docker - Academic data stays on your machine
  • Zero API costs - No LlamaParse or external PDF processing fees
  • Research data security - Everything processed and stored locally, perfect for sensitive research

Quick Start

Prerequisites

  • Python 3.11+ (required for FastMCP)
  • Docker (for Neo4j database and GROBID service)
  • Claude Desktop (recommended MCP client for academic research)

Installation

# 1. Clone repository
git clone <repository-url>
cd simple-knowledge-graph-mcp

# 2. Complete setup (installs UV, Python 3.11, academic dependencies)
./scripts/setup.sh

# 3. Start Neo4j database + GROBID academic service
./scripts/start_services.sh
docker run -d --name grobid -p 8070:8070 lfoppiano/grobid:0.8.0

# 4. Verify academic system is ready
./scripts/check_status.sh
curl http://localhost:8070/api/isalive  # Should return "true"

Configure Claude Desktop

Recommended: Easy GUI Setup
# Start HTTP server for academic tools
./scripts/start_http_server.sh

Then in Claude Desktop:

  • Settings → Connectors → Add custom connector
  • Name: Academic Knowledge Graph
  • URL: http://localhost:3001
  • Click "Add" and restart Claude Desktop
Alternative: JSON Configuration (Advanced)

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "academic-knowledge-graph": {
      "command": "uv",
      "args": ["run", "python", "/full/path/to/project/src/server/main.py"]
    }
  }
}

Start Academic Research

# Start the academic MCP server
./scripts/start_http_server.sh

# In Claude Desktop:
# 1. Upload research PDFs to your project
# 2. "Process these papers with GROBID for academic extraction"  
# 3. "Extract author networks and citation relationships"
# 4. "What are the key research trends in these papers?"
# 5. "Generate a literature review on [specific topic]"

Test Academic Processing

# Test GROBID + academic pipeline
python process_literature.py

# Your academic research assistant is now ready with:
# - GROBID's 87-90% accuracy for research papers
# - Zero API costs for PDF processing  
# - Citation network analysis
# - Author collaboration mapping
# - Academic literature review generation

That's it! Your GROBID-powered academic research assistant is ready to transform research papers into intelligent, queryable knowledge graphs.

What Makes This Academic System Superior

Unlike generic document tools or expensive PDF processing services, this system:

  1. Academic specialization - GROBID provides 87-90% F1-score accuracy specifically for research papers
  2. Zero API costs - Self-hosted GROBID eliminates LlamaParse fees ($10-30/month saved)
  3. Citation network intelligence - Maps author collaborations and paper relationships automatically
  4. Research privacy - All processing stays local, perfect for sensitive academic research
  5. Institutional analysis - Tracks university collaborations and research partnerships
  6. Academic writing support - Generates properly formatted literature reviews with citations
  7. Claude Desktop integration - Seamless research workflow within your existing environment

Perfect for researchers, graduate students, and academic institutions who need sophisticated analysis of research literature without compromising privacy or incurring API costs.