A Python agentic framework for advancing biomedical research using tools like Ollama, BioMCP, and other MCP (Model Context Protocol) servers. This framework facilitates the creation of interactive systems that utilize multiple MCP tools from various sources to assist researchers in their biomedical investigations.
- 🔌 Multi-MCP Integration: Connect to multiple MCP servers simultaneously
- 🤖 LLM Support: Integration with Ollama, OpenAI, and Anthropic models
- 📊 Research Context: Specialized data models for biomedical research contexts
- ⚡ Async Operations: High-performance asynchronous tool calls and interactions
- 💬 Interactive Sessions: Rich CLI interface with conversation management
- 📁 Export Capabilities: Export conversations in JSON and Markdown formats
- 🔧 Tool Discovery: Automatic discovery and registration of available tools
- 📈 Progress Analysis: Track and analyze research progress over time
# Clone the repository
git clone https://github.com/yourusername/biomedical-agent-framework
cd biomedical-agent-framework
# Install with all dependencies
pip install -e .[dev,ollama,biomcp]
# Or install basic version
pip install -e .-
Initialize configuration:
bioagent init-config --output my_config.yaml
-
Edit the configuration file to match your setup:
agent: model_provider: "ollama" model_name: "llama3.1" enabled_servers: - "biomcp-pubmed" mcp_servers: biomcp-pubmed: name: "BioMCP PubMed Server" endpoint: "http://localhost:8001"
Interactive Mode:
bioagent interactive --config my_config.yamlSingle Query:
bioagent query "What are the genetic factors in Alzheimer's disease?" --config my_config.yamlTest Server Connection:
bioagent test-server biomcp-pubmed --config my_config.yamlimport asyncio
from bioagent import BiomedicalAgent
from bioagent.core.models import AgentConfiguration, MCPServer, ResearchContext
async def main():
# Configure the agent
config = AgentConfiguration(
model_provider="ollama",
model_name="llama3.1",
research_context=ResearchContext(
domain="genomics",
organism="human",
research_question="What are the genetic factors in Alzheimer's disease?"
)
)
# Create and initialize agent
agent = BiomedicalAgent(config)
await agent.initialize()
# Register MCP servers
server = MCPServer(
id="biomcp-pubmed",
name="BioMCP PubMed Server",
endpoint="http://localhost:8001",
description="Access to PubMed literature database"
)
agent.register_mcp_server(server)
# Connect to servers
await agent.connect_mcp_servers()
# Start research session
session = await agent.start_session()
# Process queries
response = await agent.process_message(
"What are the latest findings about APOE4 and Alzheimer's disease?"
)
print("🤖 Response:", response.content)
if response.tool_calls:
print("🔧 Tools used:", [call.tool_name for call in response.tool_calls])
# Export conversation
conversation = session.export_conversation("markdown")
with open("research_session.md", "w") as f:
f.write(conversation)
await agent.shutdown()
if __name__ == "__main__":
asyncio.run(main())# Set research context dynamically
context = ResearchContext(
domain="drug_discovery",
organism="human",
research_question="What are potential drug targets for COVID-19?",
keywords=["covid-19", "drug targets", "antivirals"]
)
session.update_context(context)
# Analyze research progress
analysis = await session.analyze_research_progress()
print(f"Duration: {analysis['duration_minutes']:.1f} minutes")
print(f"Tools used: {analysis['tool_usage']}")
print(f"Research themes: {analysis['research_themes']}")
# Search for specific tools
tools = agent.mcp_registry.search_tools("pubmed")
for tool in tools:
print(f"- {tool.name}: {tool.description}")- Genomics: Gene expression analysis, GWAS, variant analysis
- Proteomics: Protein structure, function, and interactions
- Drug Discovery: Compound screening, target identification, ADMET
- Clinical Research: Patient data analysis, biomarker discovery
- Systems Biology: Pathway analysis, network modeling
- Bioinformatics: Sequence analysis, phylogenetics, comparative genomics
The framework is designed to work with various MCP servers:
- BioMCP Servers: Specialized biomedical databases and tools
- Ollama Tools: Local computational tools and models
- Custom Servers: Build your own MCP servers for specific needs
- PubMed (literature search)
- UniProt (protein data)
- ChEMBL (chemical/bioactivity data)
- Ensembl (genomic annotations)
- And many more through MCP servers...
# Get help
bioagent --help
# Initialize configuration
bioagent init-config
# Start interactive session
bioagent interactive --config config.yaml
# Send single query
bioagent query "your question" --config config.yaml
# List configured servers
bioagent list-servers --config config.yaml
# Test server connection
bioagent test-server server-id --config config.yaml# Clone and install in development mode
git clone https://github.com/yourusername/biomedical-agent-framework
cd biomedical-agent-framework
pip install -e .[dev]
# Run tests
pytest tests/
# Run linting
black src/
isort src/
flake8 src/biomedical-agent-framework/
├── src/bioagent/
│ ├── core/ # Core agent logic
│ ├── mcp/ # MCP client and registry
│ ├── ui/ # User interface components
│ └── cli.py # Command-line interface
├── examples/ # Usage examples
├── tests/ # Test suite
├── config/ # Configuration files
└── docs/ # Documentation
We welcome contributions! Please see our contributing guidelines and feel free to:
- Report bugs and request features
- Submit pull requests
- Improve documentation
- Add new MCP server integrations
This project is licensed under the MIT License - see the LICENSE file for details.
If you use this framework in your research, please cite:
@software{GENESIS,
title={GENESIS: An agentic framework for advancing biomedical research with MCP Integration},
author={Madduri et al.},
year={2025},
url={https://github.com/madduri/genesis}
}