Verification-Enforced Research Infrastructure with Tracking and Automated Structuring
A Claude Code research framework optimized for biomedical and medical research that enforces PubMed citation compliance, validates scientific claims in real-time, and automatically structures your knowledge base for grant applications and publications.
- Demo
- About VERITAS
- What Makes VERITAS Different
- System Architecture
- Core Features
- System Components
- Documentation
- Example Workflow
- Requirements
- Installation Details
- Creating Your Domain Configuration
- Troubleshooting
- Contributing
- License
- Acknowledgments
- Get VERITAS
VERITAS embodies the core principle of truth in biomedical research. Designed specifically for medical and life science researchers, it ensures every scientific claim is backed by PubMed citations, maintains research integrity, and streamlines grant application workflows.
- Verification-Enforced - Automatic PubMed citation verification (35+ million articles)
- Research Infrastructure - Complete framework for NIH grant applications and clinical research
- Tracking and Automated Structuring - Intelligent organization of biomedical literature and findings
GitHub: https://github.com/VMWM/VERITAS
Primary Focus: Biomedical, medical, and clinical research with PMID-based citation enforcement
Unlike other research tools, VERITAS provides:
- Real-time Citation Enforcement - PMID verification through PubMed MCP integration
- Intelligent Tool Usage - Optimized parallel/sequential execution based on operation type
- Domain Expert System - Customizable templates for any research field
- Integrated Knowledge Graph - Automatic knowledge organization via Memory MCP and structured templates
- Professional Templates - Grant-ready research questions and concept notes out of the box
- Task Tracking - TodoWrite integration for managing complex research workflows
User Message → Claude Code reads CLAUDE.md → Follows instructions → Tool Calls → Response
↓ ↓ ↓
(Constitutional rules) (Domain expert file) (MCP servers)
↓ ↓ ↓
(Output style config) (Templates & guidance) (Citations, Memory, etc.)
This architecture leverages Claude Code's built-in configuration system to ensure research integrity.
- Literature Reviews - Automatic PubMed citation verification (35+ million articles)
- Research Questions - Structured templates for grant applications
- Concept Notes - Wiki-linked knowledge base with validation
- Daily Journals - Automatic progress tracking and summarization
- PMID Verification System - Zero-tolerance policy for citation errors
- Automatic verification against PubMed database using MCP tools
- Validation of author names and publication years
- All verifications available through verification scripts
- Citation Validation - Every claim must include (Author et al., Year, PMID: XXXXXXXX)
- Format Guidelines - Tables, headers, and links follow strict standards
- Smart Execution - Prevents API errors by sequencing heavy operations (PDFs, large files)
- Memory MCP - Persistent knowledge graph for entities and relationships
- Obsidian Vaults - Direct integration with your research notes
- Session Tracking - Automatic conversation logging with 5-day retention
- PubMed (@ncukondo/pubmed-mcp) - Citation search and verification (requires NCBI credentials)
- Obsidian REST - Vault operations (supports multiple vault configurations)
- Sequential Thinking - Task decomposition and planning
- Memory - Knowledge graph storage
- Filesystem - Project file access
- Claude Transcript Reader - Session tracking and journal generation
- Special: Custom-built server that runs from VERITAS directory
- Not copied: Stays in
/claude-transcript-reader/, acts as shared service - Uses absolute path: Points to
~/VERITAS/claude-transcript-reader/index.js - Important: If you move VERITAS after installation, update the path in Claude Desktop config
- Works with built-in logs: Reads Claude Code's native JSONL conversation files
- CLAUDE.md - Constitutional rules enforced through Claude Code's output style
- Domain Expert Files - Field-specific templates and guidance in
.claude/agents/ - Slash Commands - Quick access to common workflows (optional)
- MCP Integration - Automatic tool availability in Claude Code VS Code extension
Once VERITAS is installed, your project will have this structure:
YourProject/
├── CLAUDE.md # Constitutional rules
├── .claude/ # VERITAS configuration
│ ├── agents/ # Domain expertise
│ │ └── domain-expert.md # Your field-specific configuration
│ ├── templates/ # Note templates
│ │ ├── research_question_template.md
│ │ ├── concept_template.md
│ │ └── daily_journal_template.md
│ ├── commands/ # Slash commands (optional)
│ │ ├── research-question.md
│ │ ├── concept-note.md
│ │ └── verify-citations.md
│ ├── scripts/ # Utility scripts
│ │ └── verify_pmids.py # PMID validation script (manual)
│ └── project.json # Project configuration (optional)
└── [Your existing files] # Your research content
The VERITAS repository remains at ~/VERITAS/ with this structure:
~/VERITAS/
├── assets/ # Images and logos
├── claude-transcript-reader/ # MCP transcript reader server
│ ├── index.js # MCP server implementation
│ ├── cleanup-old-transcripts.sh
│ └── install-cleanup.sh # LaunchAgent installer
├── docs/ # Documentation
│ ├── getting-started.md
│ ├── configuration-guide.md
│ └── TROUBLESHOOTING.md
├── install/
│ ├── CLAUDE.md # Constitutional document template
│ ├── scripts/ # Setup and configuration scripts
│ └── templates/
│ ├── agents/ # Domain expert templates
│ ├── commands/ # Slash command examples
│ └── obsidian/ # Note templates
└── tests/ # Verification scripts
- Installation Guide - Complete setup walkthrough with commands and troubleshooting
- Functional Tests - Verify your setup
- Configuration Guide - Adapt for your research domain and connect Obsidian
- Domain Expert Templates - Field-specific configurations
- Troubleshooting Guide - Common issues and solutions
- MCP Server Reference - Technical specifications
You: "Create a research question about machine learning in medical diagnosis"
VERITAS workflow:
1. Claude reads domain expert file for templates
2. Searches PubMed for relevant literature with verified PMIDs
3. Creates formatted note in Research Questions/
4. Generates linked concept pages
5. Stores key entities in Memory MCP
6. Tracks progress with TodoWrite
The CLAUDE.md constitutional rules and domain templates ensure consistency.
# How Does Machine Learning Improve Medical Diagnosis
## Direct Answer
Machine learning significantly enhances diagnostic accuracy...
(Smith et al., 2023, PMID: 37654321) [ABSTRACT-VERIFIED]
## Key Findings
- 92% accuracy improvement in radiology (Jones et al., 2024, PMID: 38765432)
- Reduced false positives by 45% (Chen et al., 2023, PMID: 37890123)
## Knowledge Gaps
- Limited validation in diverse populations
- Interpretability challenges remain unresolved
## Grant Applications
**Significance**: Addresses critical diagnostic errors...
**Innovation**: Novel approach combining...
**Approach**: Three-phase validation study...
## References
1. Smith J, et al. (2023). Machine Learning in Medicine. PMID: 37654321
2. Jones K, et al. (2024). AI Radiology Applications. PMID: 38765432
## Related Concepts
- [[Machine_Learning]]
- [[Medical_Diagnosis]]
- [[Diagnostic_Accuracy]]- macOS or Linux (Unix-based system required)
- Windows: Not currently supported (would require WSL)
- Claude Desktop (Claude Code) with API access
- Note: Claude Code works in both Desktop app and VS Code extension
- VS Code extension: MCP servers configured via
~/.claude.jsonper-project - Desktop app: MCP servers configured via
claude_desktop_config.json
- Node.js v16+ and npm
- Python 3.8+
- Git
- Bash shell (for scripts)
- Obsidian with Local REST API plugin (optional but recommended)
# Verify prerequisites
command -v node && echo "Node.js: OK" || echo "Node.js: MISSING"
command -v python3 && echo "Python: OK" || echo "Python: MISSING"
command -v git && echo "Git: OK" || echo "Git: MISSING"Full installation guide: docs/getting-started.md
The setup script (install/scripts/setup.sh) provides:
- Installation of VERITAS hooks and constitution
- Project type selection (medical research, software, documentation)
- Creation of
.claude/directory structure in your project - Template installation for your domain
- Conversation logger setup (special: runs from VERITAS as shared service)
- MCP server configuration for Claude Desktop
VERITAS is optimized for biomedical and medical research, with built-in support for:
- PubMed citation verification (35+ million articles)
- PMID enforcement for all scientific claims
- NIH grant application formatting
- Clinical research documentation standards
- HLA/immunology research templates
The setup script installs a medical research template (HLA antibody research). To adapt it for your specific biomedical field:
- Edit the domain expert file at
.claude/agents/hla-research-director.md - Update research focus areas to match your field (e.g., oncology, neuroscience, genetics)
- Modify grant sections for your funding agency (NIH, NSF, DoD)
- Adjust templates for your documentation needs
Ask Claude: "Adapt the HLA domain expert for [your biomedical field]:
- Research focus: [e.g., cancer immunotherapy, neurodegeneration]
- Grant type: [e.g., R01, K99/R00, F32]
- Specific requirements: [e.g., clinical trial protocols, biomarker analysis]"
VERITAS's core features assume biomedical research workflows. Adapting for other domains (software engineering, social sciences) requires manual configuration:
See the section "Adapting VERITAS for Other Domains" below for instructions.
| Issue | Solution |
|---|---|
| Claude doesn't recognize CLAUDE.md | Restart Claude Desktop after setup |
| "Command not found: claude" | Claude CLI is optional, use Desktop app |
/mcp shows no servers in VS Code |
MCP servers configured per-project in ~/.claude.json, not globally |
| MCP servers in Desktop but not VS Code | Edit ~/.claude.json and add servers to your project's mcpServers object |
| Permission prompts for every MCP tool | Add autoApprove array to ~/.claude.json - see getting-started.md |
| Validation hooks not running | Check you're in project directory with CLAUDE.md |
| Templates not found | Verify .claude/agents/domain-expert.md exists in your project |
| Obsidian connection fails | Enable HTTPS (not HTTP) in Local REST API plugin |
- Check Troubleshooting Guide
- Review Functional Tests
- Open an issue on GitHub
While VERITAS is optimized for biomedical research, it can be manually adapted for other fields. Here's what you'll need to modify:
-
Disable PubMed enforcement in
.claude/project.json:"strict_citations": false, "type": "software"
-
Create a new domain expert:
- Remove biomedical-specific sections from the HLA template
- Replace PMID requirements with GitHub/DOI citations
- Add software-specific templates (code reviews, documentation, testing)
-
Adjust MCP servers (optional):
- You may not need the PubMed server
- Consider adding GitHub or GitLab integration tools
-
Modify hooks in
.claude/hooks/:- Edit
post-command.pyto check for different citation formats - Adjust validation rules for your documentation standards
- Edit
-
Update citation style in the domain expert:
- Replace PMID with DOI/ISBN/URL citations
- Use APA/MLA/Chicago formatting
-
Modify templates for qualitative research:
- Interview protocols
- Literature review structures
- Theoretical frameworks
-
Adjust enforcement levels to match your field's standards
- Run the standard installation first
- Edit
.claude/project.jsonto change project type - Create a custom domain expert file
- Test and iterate on the configuration
Note: Core VERITAS features (sequential thinking, memory management, conversation logging) work across all domains. The main adaptations involve citation styles and domain-specific templates.
We welcome contributions!
Check out our open issues for specific tasks that need help.
- Domain Templates - Share your field-specific configurations
- MCP Integrations - Connect additional research tools
- Documentation - Improve guides and examples
- Bug Reports - Help us identify and fix issues
- Feature Requests - Suggest improvements via issues
- Check existing issues first
- Fork the repository
- Create a feature branch
- Submit a pull request to: https://github.com/VMWM/VERITAS
MIT License - See license file for details
- Built for the Claude Code research community
- Powered by MCP (Model Context Protocol) by Anthropic
- Inspired by scientific best practices
For Biomedical Research Projects: modify the bracketed sections and paste into your Claude Code chat.
Install VERITAS for me by executing these steps:
1. Clone https://github.com/VMWM/VERITAS.git to ~/VERITAS
2. Create my project directory at [REPLACE: ~/Projects/YourResearchProject]
3. Copy VERITAS core files to my project:
- CLAUDE.md (constitutional rules)
- .claude/agents/ (domain expert templates)
- .claude/templates/ (note templates)
- .claude/commands/ (slash commands - optional)
- .claude/scripts/ (utility scripts)
4. Customize the domain expert for my research area:
- Copy ~/VERITAS/install/templates/agents/hla-research-director.md
- Adapt for my field: [REPLACE: e.g., "cancer immunotherapy" or "neurodegeneration"]
- Update grant type: [REPLACE: e.g., "F31", "R01", "K99/R00"]
- Save as .claude/agents/domain-expert.md in my project
5. Install transcript cleanup (optional but recommended):
- cd ~/VERITAS/claude-transcript-reader
- Run: ./install-cleanup.sh
- This creates a LaunchAgent that runs daily at 2 AM to delete old JSONL files
6. Test the installation:
- Open my project in Claude Code
- Ask: "What is your role according to CLAUDE.md?"
- Verify VERITAS rules are recognized
Note for Claude Desktop users: MCP servers require manual configuration in claude_desktop_config.json
Note for VS Code extension users: MCP servers are automatically available, no configuration needed
Biomedical Research Installation (copy and customize):
Install VERITAS for me by executing these steps:
1. Clone https://github.com/VMWM/VERITAS.git to ~/VERITAS
2. Create my project directory at ~/Projects/MyBiomedicalResearch
3. Copy VERITAS core files to my project:
- CLAUDE.md (constitutional rules)
- .claude/agents/ (domain expert templates)
- .claude/templates/ (note templates)
- .claude/commands/ (slash commands for quick workflows)
- .claude/scripts/ (utility scripts like verify_pmids.py)
4. Install the medical research domain expert:
- Copy ~/VERITAS/install/templates/agents/hla-research-director.md
- This is pre-configured for biomedical research with PMID enforcement
- Customize research focus for my area (cancer, neuroscience, immunology, etc.)
- Update grant sections for my funding agency (NIH F31, R01, K99/R00, etc.)
- Save as .claude/agents/domain-expert.md in my project
5. Install transcript cleanup (recommended):
- Run: cd ~/VERITAS/claude-transcript-reader && ./install-cleanup.sh
- This creates a macOS LaunchAgent that runs daily at 2 AM
- Deletes JSONL transcripts older than 5 days
- Claude Code automatically logs conversations - this just manages retention
6. Test the installation:
- Open my project in Claude Code
- Ask: "What is your role according to CLAUDE.md?"
- Try: "Search PubMed for papers on immunotherapy"
- Verify VERITAS constitutional rules are being followed
For Claude Desktop users: You'll also need to configure MCP servers manually
For VS Code extension users: MCP servers work automatically
After installation: Configure Obsidian if you want vault integration (see Configuration Guide)
For interactive installation with prompts:
# Clone the repository
git clone https://github.com/VMWM/VERITAS.git
cd VERITAS
# Run setup (you'll be prompted for your project directory)
./install/scripts/setup.sh
# Configure Claude Desktop
./install/scripts/configure-claude.sh
# Restart Claude Desktop completelyNote: The setup scripts are interactive and will prompt you for choices. Claude Code cannot respond to these prompts automatically, which is why Option 1 has Claude execute the steps directly.
Quick Check: Start a new Claude Code conversation and say:
"What is your role according to CLAUDE.md?"
Claude should respond with awareness of the VERITAS system and its research integrity requirements.
Comprehensive Testing:
Run the verification scripts in the /tests folder:
# Run automated verification
cd ~/VERITAS/tests
./verify-installation.sh
# Or follow the functional test guide
cat veritas-functional-test.md
# Then test each prompt in Claude CodeThe functional tests verify:
- MCP server connections
- Citation enforcement
- Template creation
- Knowledge graph storage
- Conversation logging
For medical research users, also test:
cat veritas-functional-test-HLA-specific.md
# Test HLA-specific workflowsAfter installation, set up Obsidian manually:
- Install Obsidian and create vault(s)
- Install Local REST API plugin from Community Plugins
- Configure authentication (Bearer token or API key)
- Set port (27124 for single vault, or 27124/27125 for dual vault)
- Update MCP configuration with your Obsidian settings
See Configuration Guide for detailed Obsidian setup instructions.
Ready to transform your research workflow? Get started with VERITAS today!