jsagir/Mindrian-Reverse-Saliant-MCP-Server

🧠 Mindrian Reverse Saliant Discovery MCP Server

Discover breakthrough cross-domain innovation opportunities through dual similarity analysis

📋 Table of Contents

• What is Reverse Salient Discovery?
• Features
• Installation
• Quick Start
• Usage Examples
• API Reference
• How It Works
• Contributing
• License

🎯 What is Reverse Salient Discovery?

A Reverse Salient is a breakthrough innovation opportunity where:

• High Structural Similarity (LSA): Papers share similar methods and techniques
• Low Semantic Similarity (BERT): Papers address different problems and domains

This HIGH differential indicates that methods from one domain can be transferred to solve problems in another domain - a powerful source of innovation!

Example

Paper A: "Quantum annealing for supply chain optimization"
Paper B: "High-throughput drug combination screening"

LSA Similarity: 0.75 (HIGH - both use optimization, combinatorial methods)
BERT Similarity: 0.12 (LOW - different domains: logistics vs pharma)
Differential: 0.63 (HUGE!)

→ Innovation Opportunity: Apply quantum annealing to drug screening!

✨ Features

Core Capabilities

• ✅ LSA (Latent Semantic Analysis): Measures structural similarity via TF-IDF + SVD
• ✅ BERT Embeddings: Measures semantic similarity via contextual embeddings
• ✅ Differential Analysis: Identifies high LSA + low BERT pairs automatically
• ✅ Sequential Thinking: Transparent reasoning at every decision point

Advanced Validation

• 🔍 Patent Database Searches: Google Patents, USPTO
• 🚀 Startup Activity Monitoring: Crunchbase, TechCrunch
• 📚 Citation Network Analysis: Google Scholar, arXiv

Data Sources

• 🌐 Tavily Web Search: Multi-source academic paper collection
• 📖 Scopus API: Direct academic database access
• 📄 CSV Import: Load existing paper collections

🚀 Installation

Prerequisites

• Python 3.11 or higher
• API Keys:
  • Tavily API (required): Get here
  • Scopus API (optional): Get here

Install Steps

# 1. Clone or download the repository
git clone https://github.com/mindrian/reverse-saliant-mcp.git
cd reverse-saliant-mcp

# 2. Create virtual environment (recommended)
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# 3. Install dependencies
pip install -r requirements.txt

# 4. Download NLTK data
python -c "import nltk; nltk.download('stopwords')"

# 5. Configure environment variables
cp .env.example .env
# Edit .env and add your API keys

Environment Setup

Create .env file:

TAVILY_API_KEY=your_tavily_api_key_here
SCOPUS_API_KEY=your_scopus_api_key_here  # Optional

🎮 Quick Start

Option 1: Run Standalone

python server.py

Option 2: Connect to Claude Desktop

Edit claude_desktop_config.json:

{
  "mcpServers": {
    "mindrian": {
      "command": "python",
      "args": ["C:\\path\\to\\server.py"],
      "env": {
        "TAVILY_API_KEY": "your_key",
        "SCOPUS_API_KEY": "your_key"
      }
    }
  }
}

Restart Claude Desktop, and you'll have access to all Mindrian tools!

📚 Usage Examples

Example 1: Automated Full Workflow

# In Claude Desktop, just ask:
"""
Use Mindrian to discover innovation opportunities between 
quantum computing and drug discovery
"""

# Behind the scenes, this executes:
result = await execute_full_workflow(
    structured_input={
        "challenge": "Quantum computing × Drug discovery innovation",
        "domains": [
            {
                "label": "Quantum Computing",
                "concepts": ["superposition", "entanglement"],
                "methods": ["quantum annealing", "VQE", "QAOA"]
            },
            {
                "label": "Drug Discovery",
                "concepts": ["molecular docking", "protein folding"],
                "methods": ["high-throughput screening", "computational chemistry"]
            }
        ]
    },
    search_queries=[
        "quantum computing optimization",
        "drug discovery screening methods",
        "quantum annealing applications",
        "combinatorial drug screening"
    ],
    validate_top_n=3
)

Example 2: Step-by-Step Manual Control

# Step 1: Initialize
result = await initialize_discovery({
    "challenge": "Find AI × Healthcare innovations",
    "domains": [...]
})
session_id = result["session_id"]

# Step 2: Collect papers
await collect_papers_tavily(
    session_id=session_id,
    search_queries=["machine learning healthcare", "AI medical diagnosis"]
)

# Step 3: Clean papers
await clean_papers(session_id)

# Step 4: Compute LSA (structural similarity)
await compute_lsa_similarity(session_id)

# Step 5: Compute BERT (semantic similarity)
await compute_bert_similarity(session_id)

# Step 6: Find reverse salients
rs_result = await find_reverse_salients(session_id, top_n=20)

# Step 7: Validate top opportunity
await validate_reverse_salient(
    session_id=session_id,
    reverse_salient_id="RS-001",
    check_patents=True,
    check_startups=True,
    check_citations=True
)

# Step 8: Develop innovation thesis
await develop_innovation_thesis(session_id, "RS-001")

# Step 9: Generate report
await generate_report(session_id, format="markdown")

Example 3: Using CSV Data

# Load papers from your own CSV file
await initialize_discovery({
    "challenge": "Analyze my research corpus",
    "domains": [...]
})

await load_papers_csv(
    session_id="...",
    csv_file_path="./data/my_papers.csv"
)

# Continue with normal workflow

🔧 API Reference

Core Tools

Input Format

structured_input = {
    "challenge": "Description of innovation challenge",
    "domains": [
        {
            "label": "Domain Name",
            "concepts": ["concept1", "concept2", "concept3"],
            "methods": ["method1", "method2"],
            "problems": ["problem1", "problem2"],
            "terminology": ["term1", "term2"]
        }
    ],
    "constraints": ["constraint1", "constraint2"],
    "metadata": {
        "industry": "pharmaceutical",
        "timeline": "2-3 years"
    }
}

🧠 How It Works

The Dual Similarity Framework

┌─────────────────────────────────────────────────────────┐
│                   PAPER COLLECTION                       │
│  Tavily Search → Scopus API → CSV Import → Cleaning     │
└────────────────────┬────────────────────────────────────┘
                     │
         ┌───────────┴───────────┐
         │                       │
         ▼                       ▼
┌────────────────┐      ┌────────────────┐
│ LSA ANALYSIS   │      │ BERT ANALYSIS  │
│ (Structural)   │      │ (Semantic)     │
├────────────────┤      ├────────────────┤
│ • TF-IDF       │      │ • Tokenization │
│ • SVD Topics   │      │ • Embeddings   │
│ • Similarity   │      │ • Cosine Sim   │
└────────┬───────┘      └───────┬────────┘
         │                      │
         └──────────┬───────────┘
                    ▼
         ┌─────────────────────┐
         │ DIFFERENTIAL        │
         │ |BERT - LSA|        │
         │                     │
         │ HIGH = Innovation!  │
         └──────────┬──────────┘
                    │
         ┌──────────┴───────────┐
         │                      │
         ▼                      ▼
┌────────────────┐    ┌────────────────┐
│ VALIDATION     │    │ INNOVATION     │
│ • Patents      │    │ THESIS         │
│ • Startups     │    │ • Mechanism    │
│ • Citations    │    │ • Feasibility  │
└────────────────┘    └────────────────┘

Sequential Thinking Integration

Every phase includes transparent reasoning:

{
  "thought_number": 1,
  "thought": "Analyzing 2 domains. Need to: (1) Analyze characteristics, 
             (2) Plan search strategy, (3) Determine data collection.",
  "next_thought_needed": True,
  "timestamp": "2025-01-15T10:30:00"
}

All thinking logs are stored and included in final reports.

📊 Understanding Results

Reverse Salient Example

{
  "id": "RS-001",
  "rank": 1,
  "lsa_similarity": 0.72,
  "bert_similarity": 0.15,
  "differential_score": 0.57,
  "breakthrough_potential": 9,
  "interpretation": "High LSA, Low BERT (INNOVATION!)"
}

Interpretation:

• LSA 0.72 = Papers use 72% similar methods
• BERT 0.15 = Papers only 15% similar in meaning
• Differential 0.57 = HUGE gap = strong innovation signal
• Potential 9/10 = Highly promising opportunity

Validation Results

{
  "checks_performed": [
    {
      "check": "patents",
      "patents_found": 2,
      "novelty_score": 8,
      "status": "Clear"
    },
    {
      "check": "startups",
      "companies_found": 1,
      "market_maturity": "Early",
      "competition_level": "Low"
    },
    {
      "check": "citations",
      "recent_papers_found": 5,
      "research_activity": "Low",
      "novelty_indicator": "High novelty"
    }
  ],
  "overall_novelty_score": 9,
  "recommendation": "HIGH PRIORITY - Novel opportunity"
}

🐛 Troubleshooting

Common Issues

"TAVILY_API_KEY not set"

• Solution: Create .env file with your API key

"No papers collected"

• Check your API key is valid
• Try broader search queries
• Verify internet connection

"BERT computation very slow"

• This is normal for 50+ papers
• Use bert-base-uncased instead of bert-large-uncased
• Consider running on GPU if available

"LSA matrix all zeros"

• Papers might be too short
• Try different max_features parameter
• Check cleaned papers aren't empty

Performance Tips

• 50 papers: ~5 minutes total
• 100 papers: ~10 minutes total
• 500 papers: ~60 minutes total
• Use CSV import for repeated runs
• Cache LSA/BERT matrices if rerunning

🤝 Contributing

Contributions welcome! Please:

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Add tests if applicable
5. Submit a pull request

📄 License

MIT License - see for details

🙏 Acknowledgments

• Original research methodology from LSA/BERT dual similarity analysis
• FastMCP framework by Anthropic
• Tavily Search API
• Elsevier Scopus API

📧 Contact

For questions, issues, or collaboration:

• GitHub Issues: github.com/mindrian/reverse-saliant-mcp/issues
• Email: contact@mindrian.com

Built with ❤️ by Mindrian Labs

Discover the innovations hiding between domains.