ResearchTwin MCP Server

ResearchTwin: Federated Agentic Web of Research Knowledge

ResearchTwin is an open-source, federated platform that transforms a researcher's publications, datasets, and code repositories into a conversational Digital Twin. Built on a Bimodal Glial-Neural Optimization (BGNO) architecture, it enables dual-discovery where both humans and AI agents collaborate to accelerate scientific discovery.

Live at researchtwin.net | Join the Network

Project Vision

The exponential growth of scientific outputs has created a "discovery bottleneck." Traditional static PDFs and siloed repositories limit knowledge synthesis and reuse. ResearchTwin addresses this by:

• Integrating multi-modal research artifacts from Semantic Scholar, Google Scholar, GitHub, and Figshare
• Computing a real-time S-Index metric (Quality × Impact × Collaboration) across all output types
• Providing a conversational chatbot interface for interactive research exploration
• Exposing an Inter-Agentic Discovery API with Schema.org types for machine-to-machine research discovery
• Enabling a federated, Discord-like architecture supporting local nodes, hubs, and hosted edges

Architecture Overview

BGNO (Bimodal Glial-Neural Optimization)

Data Sources          Glial Layer          Neural Layer         Interface
┌──────────────┐    ┌─────────────┐    ┌──────────────┐    ┌────────────┐
│Semantic Scholar│───▶│             │    │              │    │  Web Chat  │
│Google Scholar │───▶│  SQLite     │───▶│  RAG with    │───▶│  Discord   │
│GitHub API     │───▶│  Cache +    │    │  Claude API  │    │  Agent API │
│Figshare API   │───▶│  Rate Limit │    │              │    │  Embed     │
└──────────────┘    └─────────────┘    └──────────────┘    └────────────┘

• Connector Layer: Pulls papers (S2+GS with deduplication), repos (GitHub), datasets (Figshare), and ORCID metadata
• Glial Layer: SQLite caching with 24h TTL, rate limiting, S2+GS title-similarity merge (0.85 threshold)
• Neural Layer: RAG with Claude — context assembly, prompt engineering, conversational synthesis
• Interface Layer: D3.js knowledge graph, chat widget, Discord bot, REST API

Federated Network Tiers

Inter-Agentic Discovery API

Machine-readable endpoints with Schema.org @type annotations:

Getting Started

Hosted (Tier 3) — Zero Setup

1. Visit researchtwin.net/join.html
2. Register with your name, email, and research identifiers
3. Your Digital Twin is live immediately

Local Node (Tier 1) — Full Control

git clone https://github.com/martinfrasch/researchtwin.git
cd researchtwin
pip install -r backend/requirements.txt
cp node_config.json.example node_config.json
# Edit node_config.json with your details
python run_node.py --config node_config.json

Docker Deployment

cp .env.example .env  # Add your API keys
docker-compose up -d --build

Required API keys: ANTHROPIC_API_KEY (for Claude RAG) Optional: S2_API_KEY, GITHUB_TOKEN, DISCORD_BOT_TOKEN, SMTP credentials

Repository Structure

researchtwin/
├── backend/
│   ├── main.py              # FastAPI endpoints (REST + Discovery API)
│   ├── researchers.py        # SQLite researcher CRUD + token management
│   ├── database.py           # SQLite schema, WAL mode, migrations
│   ├── models.py             # Pydantic models for all endpoints
│   ├── rag.py                # RAG context assembly for Claude
│   ├── qic_index.py          # S-Index / QIC computation engine
│   ├