科学论文验证分布式服务器

n8n Workflow → Submit Papers → MCP Server → Create Jobs
                                    ↓
              Android Devices ← Fetch Jobs ← SQLite DB
                    ↓
            Validate Sections → Submit Results → Consensus
                                                    ↓
                              Unity Game ← SSE Updates ← Leaderboard

┌─────────────────────────────────────────────────────────────────────┐
│                    PaperStream MCP Server v1.0                       │
├─────────────────────────────────────────────────────────────────────┤
│                                                                      │
│  ┌─────────────┐    ┌─────────────┐    ┌─────────────┐              │
│  │  REST API   │    │   MCP API   │    │  Unity SSE  │              │
│  │  /api/*     │    │   /mcp      │    │  /stream    │              │
│  └──────┬──────┘    └──────┬──────┘    └──────┬──────┘              │
│         │                  │                   │                     │
│  ┌──────┴──────────────────┴───────────────────┴──────┐             │
│  │                   Starlette App                     │             │
│  └─────────────────────────┬───────────────────────────┘             │
│                            │                                         │
│  ┌─────────────────────────┴───────────────────────────┐             │
│  │                     Handlers                         │             │
│  │  ┌──────────┐  ┌──────────┐  ┌──────────┐  ┌──────┐ │             │
│  │  │  Paper   │  │   Rule   │  │   Job    │  │Device│ │             │
│  │  │ Handler  │  │ Handler  │  │ Handler  │  │Handler│             │
│  │  └────┬─────┘  └────┬─────┘  └────┬─────┘  └───┬───┘ │             │
│  └───────┼─────────────┼─────────────┼────────────┼─────┘             │
│          │             │             │            │                   │
│  ┌───────┴─────────────┴─────────────┴────────────┴─────┐             │
│  │                    SQLite Database                    │             │
│  │  papers | rules | jobs | devices | results | consensus│             │
│  └───────────────────────────────────────────────────────┘             │
│                                                                      │
│  ┌─────────────────────────┐    ┌─────────────────────────┐          │
│  │      BioBERT Handler    │    │   Paper Processor       │          │
│  │   (Embeddings, 768-dim) │    │ (PDF → Sections → Voxels)│          │
│  └─────────────────────────┘    └─────────────────────────┘          │
└─────────────────────────────────────────────────────────────────────┘

# Clone repository
git clone https://github.com/Nileneb/mcp-paperstream.git
cd mcp-paperstream

# Create virtual environment
python -m venv .venv
source .venv/bin/activate

# Install dependencies
pip install torch transformers fastmcp pyyaml httpx pillow python-dotenv aiohttp uvicorn starlette requests

# Optional: PDF processing
pip install PyMuPDF

# Download models (auto-downloads from HuggingFace on first use)
python -m src.paperstream.handlers.download_model all

# Integrated server (recommended)
./start_server.sh

# Or manually:
python -m uvicorn src.paperstream.server_integrated:app --host 0.0.0.0 --port 8089

# MCP-only mode (BERTScore IoT):
./start_server.sh mcp

# Health check
curl http://localhost:8089/health

# Submit paper (REST API)
curl -X POST http://localhost:8089/api/papers/submit \
  -H "Content-Type: application/json" \
  -d '{"paper_id": "PMC12345", "title": "My Paper", "priority": 7}'

# Get stats
curl http://localhost:8089/api/stats

{
  "name": "submit_paper",
  "arguments": {
    "paper_id": "PMC12345",
    "title": "My Scientific Paper",
    "pdf_url": "https://example.com/paper.pdf",
    "priority": 8,
    "source": "n8n"
  }
}

{
  "name": "create_rule",
  "arguments": {
    "rule_id": "is_rct",
    "question": "Is this a randomized controlled trial?",
    "positive_phrases": ["randomized controlled trial", "RCT", "clinical trial"],
    "negative_phrases": ["review", "meta-analysis"],
    "threshold": 0.75
  }
}

src/paperstream/
├── server_integrated.py   # Main server (REST + MCP + SSE)
├── server.py              # MCP-only server (BERTScore IoT)
├── config.yaml            # Central configuration
│
├── db/                    # Database Layer
│   ├── database.py        # SQLite manager
│   ├── models.py          # Dataclass models
│   └── migrations/        # SQL schema
│
├── api/                   # API Handlers
│   ├── paper_handler.py   # Paper CRUD
│   ├── rule_handler.py    # Rule management + BioBERT
│   ├── job_handler.py     # Job distribution
│   ├── device_handler.py  # Device registration
│   └── sse_stream.py      # Unity SSE stream
│
├── pipeline/              # Processing Pipeline
│   ├── paper_processor.py # PDF → Sections → Voxels
│   └── consensus_engine.py# Result aggregation
│
├── handlers/              # ML Handlers
│   ├── biobert_handler.py # BioBERT embeddings
│   ├── sd_api_client.py   # Stable Diffusion API
│   └── biomedclip_handler.py # BiomedCLIP (optional)
│
└── prompts/               # Prompt Templates
    ├── scientific_templates.py
    └── term_mappings.json

server:
  host: "0.0.0.0"
  port: 8089

models:
  biobert:
    name: "nlpie/distil-biobert"
    path: "./src/paperstream/models/biobert"

consensus:
  min_votes: 3
  agreement_threshold: 0.6

jobs:
  ttl_seconds: 300
  max_per_device: 5

# Test all modules
python -c "
from src.paperstream.db import get_db
from src.paperstream.api import get_paper_handler, get_rule_handler
from src.paperstream.pipeline import get_consensus_engine
print('All modules OK')
"

# Initialize database
python -c "
from src.paperstream.db import get_db
db = get_db()
db.initialize()
print(f'DB: {db.db_path}')
"

# Load default rules
curl http://localhost:8089/api/stats