MCP Servers
External tool integrations
MCP (Model Context Protocol) servers connect Claude to external tools, databases, and services.
What Are MCP Servers?
MCP servers provide Claude with: - Access to databases (PubMed, UniProt, etc.) - Integration with lab systems - Specialized scientific tools - Data visualization capabilities
Recommended Plugins
The lab recommends these plugins (which include MCP servers):
scientific-skills
Scientific databases and analysis tools:
claude plugins install scientific-skills@claude-scientific-skillsProvides access to: - Databases: PubMed, UniProt, KEGG, ChEMBL, ClinicalTrials.gov - Analysis: Statistical methods, visualization - Data formats: FASTA, PDB, CSV, Excel
superpowers
Development workflow tools:
claude plugins install superpowers@superpowers-marketplaceProvides: - Code review agents - TDD workflows - Debugging tools - Git worktree management
github
GitHub CLI integration:
claude plugins install github@claude-plugins-officialProvides: - Issue management - Pull request workflows - Repository operations
Using MCP-Provided Tools
Database Queries
With scientific-skills installed:
> Search PubMed for recent papers on TREM2 and Alzheimer's
> Get the protein sequence for human TP53 from UniProt
> What clinical trials are recruiting for glioblastoma?
Data Analysis
> Perform a power analysis for a two-sample t-test
> Create a volcano plot from this differential expression data
> Run a survival analysis on the patient data
File Operations
> Read the Excel file and summarize the data
> Convert this FASTA to a table format
> Parse the PDB file and extract chain A coordinates
Configuration
MCP servers are configured through plugins. Check your configuration:
# List installed plugins
claude plugins list
# Check what tools are available
claude tools listLab-Specific Integrations
Accessing Lab Data
For projects that need lab-specific integrations:
{
"mcp": {
"servers": {
"lab-data": {
"command": "python",
"args": ["scripts/lab_data_server.py"],
"description": "Access to /proj/rashidlab/ data"
}
}
}
}Custom Scientific Tools
For specialized analysis:
{
"mcp": {
"servers": {
"simulation-tools": {
"command": "Rscript",
"args": ["scripts/simulation_server.R"],
"description": "Run simulations with lab parameters"
}
}
}
}Available Skills from Plugins
scientific-skills Database Tools
| Tool | Description |
|---|---|
| pubmed-database | Search PubMed literature |
| uniprot-database | Query protein sequences |
| pdb-database | Access protein structures |
| kegg-database | Query pathway data |
| chembl-database | Search compound data |
| clinicaltrials-database | Find clinical trials |
scientific-skills Analysis Tools
| Tool | Description |
|---|---|
| statistical-analysis | Statistical tests |
| exploratory-data-analysis | EDA workflows |
| scientific-visualization | Create figures |
| hypothesis-generation | Research planning |
superpowers Development Tools
| Tool | Description |
|---|---|
| code-review | Review code changes |
| test-driven-development | TDD workflows |
| systematic-debugging | Debug issues |
| brainstorming | Design discussions |
Example Workflows
Literature Search
> Search PubMed for papers on Bayesian adaptive trials
> published in the last 2 years
Claude uses pubmed-database to query and summarize results.
Protein Analysis
> Get the TP53 sequence from UniProt and identify
> functional domains
Claude fetches data and analyzes the protein.
Statistical Analysis
> Run a power analysis for detecting a 0.5 effect size
> with alpha=0.05
Claude uses statistical tools to compute power.
Troubleshooting
Tool not found
# Reinstall the plugin
claude plugins uninstall scientific-skills@claude-scientific-skills
claude plugins install scientific-skills@claude-scientific-skillsServer connection failed
# Check plugin status
claude plugins list
# Check for updates
claude plugins updateTool runs slowly
Large database queries may take time. For complex queries, consider: - Breaking into smaller queries - Using more specific search terms - Caching results locally
Security Notes
- MCP servers can access external resources
- Be cautious with sensitive data
- Lab data should stay on approved systems
- Don’t expose credentials in MCP configurations