MedCP transforms Claude Desktop into a powerful medical AI assistant by providing secure, local access to electronic health records and biomedical knowledge graphs. Process sensitive health data entirely on your machine while delivering instant access to clinical insights.
- Local Processing - All data processing happens on your machine
- EHR Integration - Query electronic health records with natural language
- Biomedical Knowledge - Access comprehensive drug-disease associations and protein interactions
- Real-time Analysis - Instant clinical decision support
- Secure Storage - Credentials encrypted in OS keychain
- Claude Desktop 1.0.0+ with MCPB extension support
- Operating System: macOS 11+ or Windows 10+
- Python Included (standalone runtime bundled with extension)
- Memory: 8GB RAM minimum, 16GB recommended
Download Claude Desktop Visit claude.ai/download and install the latest version.
That's all! Python and all dependencies are bundled with the extension.
-
Download the Extension
- Go to Releases
- Download the latest
MedCP.mcpbfile
-
Install in Claude Desktop
- Double-click the
MedCP.mcpbfile - Claude Desktop will open the installation dialog
- Click "Install"
- Double-click the
-
Configure Databases
- Complete the configuration wizard that appears
- Enter your database credentials (details below)
That's it!
Run MedCP directly from the repository using uvx without installation:
# Required: Set your credentials as environment variables
# At least ONE of the following database configurations must be provided:
# Option A: Knowledge Graph only (biomedical knowledge inference)
export KNOWLEDGE_GRAPH_URI="bolt://your-neo4j-server:7687"
export KNOWLEDGE_GRAPH_USERNAME="your_username"
export KNOWLEDGE_GRAPH_PASSWORD="your_password"
export KNOWLEDGE_GRAPH_DATABASE="neo4j"
# Option B: Clinical Records only (EHR queries)
export CLINICAL_RECORDS_SERVER="your-server.hospital.org"
export CLINICAL_RECORDS_DATABASE="your_database"
export CLINICAL_RECORDS_USERNAME="your_username"
export CLINICAL_RECORDS_PASSWORD="your_password"
# Option C: Both (for integrated medical analysis)
# Set all environment variables above
# Run from GitHub
uvx --from git+https://github.com/BaranziniLab/MedCP medcpImportant Notes:
-
If only Knowledge Graph is configured, you'll have access to:
get_knowledge_graph_schema- List all biomedical entities and relationshipsquery_knowledge_graph- Query drug-disease associations, protein interactions, etc.
-
If only Clinical Records is configured, you'll have access to:
list_clinical_tables- List available EHR tablesquery_clinical_records- Query patient records with SQL
-
If both are configured, you'll have access to all tools for integrated analysis
Or run from a local clone:
# Clone the repository
git clone https://github.com/BaranziniLab/MedCP.git
cd MedCP
# Set environment variables (see .env.example)
export KNOWLEDGE_GRAPH_URI="bolt://localhost:7687"
# ... set other variables ...
# Run with uvx
uvx --from . medcpAfter installation, you'll need to configure your database connections in Claude Desktop:
Settings → Extensions → MedCP
MedCP uses the SPOKE knowledge graph by default (Morris et al., 2023), which contains comprehensive biomedical relationships including drug-disease associations, protein interactions, and biological pathways.
| Parameter | Description | Example |
|---|---|---|
| Knowledge Graph URI | Neo4j connection URI | bolt://your-neo4j-server:7687 |
| Username | Neo4j database username | your_username |
| Password | Neo4j database password | your_secure_password |
| Database Name | Neo4j database name | spoke (default) |
Configure access to your clinical database. For UCSF users, see the UCSF Research Data portal for access information.
| Parameter | Description | Example |
|---|---|---|
| Clinical Records Server | SQL Server hostname | your-ehr-server.hospital.org |
| Database Name | Clinical database name | OMOP_DEID |
| Username | Database username | clinical_user |
| Password | Database password | secure_clinical_password |
| Parameter | Description | Default |
|---|---|---|
| MedCP Namespace | Tool prefix for organization | MedCP |
| Log Level | Logging verbosity | INFO |
Security Note: All sensitive credentials are automatically encrypted and stored in your operating system's secure keychain.
"Find all patients diagnosed with diabetes in the last 6 months and summarize their HbA1c trends"
"Check for interactions between metformin, lisinopril, and atorvastatin for a 65-year-old patient with CKD stage 3"
"What are the current evidence-based guidelines for treating community-acquired pneumonia in elderly patients?"
"Find protein targets associated with Alzheimer's disease and identify potential drug compounds that interact with these proteins"
- Verify Claude Desktop supports MCPB extensions
- Check that all required configuration fields are completed
- Restart Claude Desktop after configuration changes
- Verify server URLs are accessible from your network
- Check that credentials are valid and not expired
- Ensure firewall allows database connections
- Test connectivity outside Claude Desktop if possible
- Limit query result sizes for large datasets
- Use specific date ranges in clinical queries
- Check available system memory
- Documentation: Check the configuration examples above
- Bug Reports: Create an issue
- Security Issues: Contact the development team privately
MedCP is released under the MIT License.
MedCP is developed and maintained by the Baranzini Lab at UCSF.
- Wanjun Gu - wanjun.gu@ucsf.edu
- Gianmarco Bellucci - gianmarco.bellucci@ucsf.edu
- SPOKE Knowledge Graph: Morris et al., 2023
- UCSF Clinical Data: UCSF Research Data Portal
- Desktop Extensions: Built using Model Context Protocol Bundle (MCPB) format
- Model Context Protocol: Enables secure local AI integration