An autonomous pipeline for converting scientific papers into executable Morpheus biological simulations using LLM-based agents with MCP (Model Context Protocol) tools.
The Morpheus Benchmark Runner enables fully autonomous processing of scientific papers to generate MorpheusML simulation models. The system uses Claude AI with MCP tools to:
- Extract text from scientific PDFs
- Consult reference XML templates
- Generate MorpheusML models
- Execute Morpheus simulations
- Self-correct errors automatically
- Evaluate and score results
No manual intervention required. Paper to simulation in ~7 minutes.
Morpheus is a modeling and simulation environment for multicellular systems biology, used for computational models of cell migration, tissue development, and pattern formation. Models are defined using MorpheusML (XML-based).
The Model Context Protocol (MCP) is a standardized interface for connecting AI models to external tools. In this project, MCP tools are Python functions that Claude calls to interact with the Morpheus environment.
┌─────────────────────────────────────────────────────────────────┐
│ MORPHEUS BENCHMARK RUNNER │
├─────────────────────────────────────────────────────────────────┤
│ │
│ ┌──────────┐ ┌──────────────┐ ┌──────────────────┐ │
│ │ LLM │────▶│ Tool Request │────▶│ MCP Functions │ │
│ │ API │◀────│ Tool Result │◀────│ (server.py) │ │
│ └──────────┘ └──────────────┘ └──────────────────┘ │
│ │ │ │
│ │ ▼ │
│ │ ┌──────────────────┐ │
│ │ │ Morpheus │ │
│ │ │ Simulator │ │
│ │ └──────────────────┘ │
│ ▼ │ │
│ ┌──────────┐ ┌──────────────────┐ │
│ │ Benchmark│ │ Output Files │ │
│ │ Results │ │ (PNG, CSV, XML) │ │
│ └──────────┘ └──────────────────┘ │
│ │
│ Loop repeats until "PAPER_COMPLETE" is signaled │
└─────────────────────────────────────────────────────────────────┘
Complete MCP tool orchestration workflow for autonomous model generation
| Feature | Description |
|---|---|
| 🤖 Autonomous Processing | Papers processed without manual intervention |
| 📚 Reference-Grounded Generation | XML models based on official Morpheus examples |
| 🔧 Automatic Error Recovery | Failed simulations diagnosed and fixed (up to 2 attempts) |
| 📊 Comprehensive Evaluation | 7-point scoring system for quality assessment |
| ⏱️ Rate Limit Handling | Automatic exponential backoff for API limits |
| 💾 JSON Output | Structured results for further analysis |
| ✅ Gnuplotter Validation | Ensures visualization output is configured |
git clone https://github.com/yourusername/Morpheus.AI.git
cd Morpheus.AI
python -m venv .venv
source .venv/bin/activate # Windows: .venv\Scripts\activate
pip install -r requirements.txtexport ANTHROPIC_API_KEY="sk-ant-api03-your-key-here"python run_benchmark.py📖 For detailed installation and configuration, see SETUP.md
- PDF Extraction →
pdf_to_morpheus_pipeline() - Reference Discovery →
list_references() - Template Loading →
read_reference() - XML Generation →
generate_xml_from_text() - Simulation Execution →
run_morpheus() - Error Recovery →
auto_fix_and_rerun()(if needed) - Evaluation →
evaluation() - Completion → "PAPER_COMPLETE" signal
Each run is scored on a 0-7 point scale:
| Criterion | Points | Description |
|---|---|---|
| XML Errors | 0 to -N | Penalty for each error in stderr.log |
| Model Graph | +1 | model_graph.dot is generated |
| Time Progression | +1 to +3 | Simulation steps (1-10: +1, 11-50: +2, 51+: +3) |
| StopTime Match | +1 | Simulation completes to configured StopTime |
| Results Generated | +1 | At least one PNG or CSV produced |
| Bonus | +1 | 10+ PNG files generated |
Score Interpretation:
- 7/7 → Perfect run
- 5-6/7 → Good run, minor issues
- 3-4/7 → Partial success
- 0-2/7 → Failed or major problems
morpheus-benchmark-runner/
├── server.py # MCP tool implementations
├── run_benchmark.py # Autonomous agent runner
├── requirements.txt # Python dependencies
├── .env # API key (optional)
├── README.md # This file
├── SETUP.md # Detailed setup guide
│
├── references/ # Reference XML templates
│ ├── CPM/ # Cellular Potts Model examples
│ ├── PDE/ # Partial Differential Equations
│ ├── ODE/ # Ordinary Differential Equations
│ ├── Multiscale/ # Multiscale models
│ └── Miscellaneous/ # Other examples
│
└── papers/ # Input PDF papers
├── paper1.pdf
└── ...
══════════════════════════════════════════════════════════════════
MORPHEUS BENCHMARK RUNNER
══════════════════════════════════════════════════════════════════
Model: claude-sonnet-4-5-20250929
Papers: 10
Max iterations per paper: 25
══════════════════════════════════════════════════════════════════
Processing paper 1/10: paper1.pdf
✓ PDF extracted
✓ References loaded
✓ XML generated
✓ Simulation completed
✓ Evaluation: 7/7
────────────────────────────────────────────────────────────────
Paper Result: COMPLETED
Score: 7/7 | PNGs: 249 | CSVs: 150 | Iterations: 23
────────────────────────────────────────────────────────────────
| Model | Est. Cost (10 papers) | Processing Time |
|---|---|---|
| claude-sonnet-4-20250514 | $5-10 | ~60-90 min |
| claude-sonnet-4-5/6-20250929 | $8-15 | ~70-100 min |
| claude-opus-4-5/6-20250514 | $15-30 | ~90-120 min |
- SETUP.md — Detailed installation, configuration, and troubleshooting
- API Reference — MCP tool documentation
The system was evaluated on 10 scientific papers spanning diverse areas of multicellular biology.
| Metric | Value |
|---|---|
| Success Rate | 90% (9/10 papers) |
| Average Score | 6.3 / 7 |
| Perfect Scores (7/7) | 9 papers |
| Total PNGs Generated | 944 |
| Total CSVs Generated | 162 |
| Total Processing Time | 70 min 39 sec |
| Avg. Time per Paper | ~7 minutes |
| # | Paper | Domain | Score | PNGs | CSVs | Iterations | Status |
|---|---|---|---|---|---|---|---|
| 1 | Szabó et al. (2010) | Cell Sorting | 7/7 | 249 | 150 | 23 | ✅ |
| 2 | Thapa et al. (2024) | Tissue Development | 7/7 | 112 | 1 | 15 | ✅ |
| 3 | Carpenter et al. (2024) | Cell Migration | 7/7 | 62 | 1 | 8 | ✅ |
| 4 | Atwell et al. (2025) | Germ Cell Proliferation | 7/7 | 111 | 1 | 12 | ✅ |
| 5 | Berkemeier et al. (2023) | Cellular Mechanics | 7/7 | 52 | 1 | 13 | ✅ |
| 6 | Herron et al. (2022) | Multicellularity | 7/7 | 43 | 2 | 10 | ✅ |
| 7 | Scheel et al. (2021) | Cell Invasion | 7/7 | 111 | 2 | 16 | ✅ |
| 8 | Mu et al. (2024) | Morphogenesis | 7/7 | 103 | 1 | 14 | ✅ |
| 9 | Luque et al. (2023) | Pattern Formation | 7/7 | 101 | 3 | 23 | ✅ |
| 10 | Berkhout et al. (2025) | Population Dynamics | 0/7 | 0 | 0 | 25 | ❌ |
Generated simulations were compared against original paper figures to assess biological accuracy:
| Original Output | Generated Output |
|---|---|
 |
 |
| The senescent mesothelial matrix accentuates colonization by ovarian cancer cells (Thapa et al., 2024) : Differential adhesion driven segregation | |
 |
 |
| multiscale spatial modeling framework for the germinal center response (Mu et al., 2024) : Concentration gradients of chemoattractants | |
| Model | Success Rate | Avg. Iterations | Processing Time | Est. Cost |
|---|---|---|---|---|
| Sonnet 4 | 70% | 18.2 | 95 min | ~$6.50 |
| Sonnet 4.5 | 90% | 15.9 | 71 min | ~$11.50 |
| Opus 4.5 | 90% | 14.7 | 68 min | ~$28.00 |
Recommendation: Sonnet 4.5 offers the best cost-performance balance.
Detailed workflow demonstrations are available:
| Demo | Description | Link |
|---|---|---|
| 🎬 Single Paper Processing | Claude Desktop with Morpheus MCP tools | Watch on YouTube |
| 🎬 10-Paper Benchmark | Autonomous API execution with MCP tools | Watch on YouTube |
MIT License — see LICENSE for details.