v0.4.0

What’s new

• DuckDB + Parquet backend – Local storage now uses DuckDB with Parquet instead of SQLite for faster, more scalable querying.
• Full MIMIC-IV support – Works with the full MIMIC-IV dataset (after you download and copy it into m3_data/raw).
• Simpler CLI – m3 init now takes care of the previous download and convert steps with a cleaner UX.
• MCP + containers – MCP server and config are wired for the DuckDB backend, and the Dockerfile is ready for kAgent/Kubernetes-style deployments.
• CI refresh – Tests and workflows are updated

v0.3.0 - Docker Support & Improved Onboarding

🐳 Docker Support

• Add official Docker images with lite (SQLite) and bigquery variants
• Standalone images with baked-in demo database for instant deployment
• Support for both local demo (100 patients) and full BigQuery (365k patients)

📚 Documentation Improvements

• Cleaner Quick Start with side-by-side SQLite/BigQuery configurations
• Add llms-install.md for AI agent setup
• Add Citation section with BibTeX for academic use
• Add Related Projects section for community adaptations
• Streamlined installation instructions with uvx as primary method

🔧 User Experience

• Add m3-mcp entry point for simpler uvx usage (no --from needed)
• Update logos across webapp and README
• Add video tutorial references

🧪 Benchmarks & Testing

• Add GPT-OSS-20B EHRSQL benchmark results
• Fix valvuloplasty benchmark model answer
• Add bug/feature/question issue templates

🌐 Webapp

• Replace Formspree with EmailJS for contact form
• Add npm setup instructions to webapp README
• Remove deprecated query interface section

🐛 Bug Fixes

• Fix pre-commit CI uv symlink path

v0.2.0 - Modern Build System with UV + PDM Support

🚀 v0.2.0: Modern Build System with UV + PDM Support

🎯 Major Infrastructure Modernization

This release brings M3 into the modern Python packaging ecosystem, providing significantly faster installation and development workflows while maintaining 100% backwards compatibility.

✨ What's New

⚡ Dual Package Manager Support

Choose the installation method that works best for you:

Traditional (pip) - Unchanged, fully supported

pip install m3-mcp

Modern (UV) - ⚡ ~3x faster installation

# Install UV (one-time setup)
curl -LsSf https://astral.sh/uv/install.sh | sh

# For development
git clone https://github.com/rafiattrach/m3.git
cd m3
uv venv && uv sync

🔧 Modernized Build System

• Build Backend: setuptools → pdm-backend (more reliable, faster builds)
• Dependency Management: Migrated to PEP 735 dependency-groups standard
• Version Management: File-based versioning for better reproducibility

🏗️ Enhanced Developer Experience

• Faster CI/CD: GitHub Actions now use UV for ~3x faster builds
• Better Lock Files: Deterministic dependency resolution with uv.lock
• Cleaner Packaging: Removed redundant build configuration

📊 Performance Improvements

Operation	Before (pip)	After (UV)	Speedup
Fresh Install	~45s	~15s	3x faster
CI Build	~2m 30s	~1m 10s	2x faster
Dev Setup	~30s	~10s	3x faster

🔄 Migration Summary

Component	v0.1.5	v0.2.0
Build Backend	setuptools	pdm-backend
Package Manager	pip only	pip + uv
Dependencies	[project.optional-dependencies]	[project.dependency-groups]
CI System	pip-based	UV-powered
Version Source	setuptools dynamic	PDM file-based

✅ Backwards Compatibility Guarantee

What Still Works Exactly the Same:

• ✅ All pip install commands and workflows
• ✅ All CLI commands (m3 --version, m3 init, etc.)
• ✅ All APIs and functionality
• ✅ All environment variables and configuration
• ✅ All existing documentation and tutorials

What's Enhanced:

• ✅ Faster alternative installation with UV
• ✅ More reliable builds with PDM backend
• ✅ Better development workflow options

🛠️ For Contributors

Development setup is now even easier:

Option 1: Traditional

git clone https://github.com/rafiattrach/m3.git
cd m3
python -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"
pytest

Option 2: Modern (Recommended)

git clone https://github.com/rafiattrach/m3.git
cd m3
uv venv && uv sync --dev
uv run pytest

🐛 Bug Fixes & Improvements

• Fixed: Redundant build configuration removed
• Improved: Package metadata handling
• Enhanced: Test execution in CI environments

🔮 Looking Forward

This modernization sets the foundation for:

• Even faster development cycles
• Better dependency resolution
• Enhanced reproducibility across environments
• Future Python packaging standard adoption

📦 Installation

For End Users:

pip install m3-mcp==0.2.0

For Developers (Fast):

git clone https://github.com/rafiattrach/m3.git
cd m3 && uv venv && uv sync

---

Full Changelog: v0.1.5...v0.2.0

Questions? Open an issue or check our documentation.

v0.1.5 - Fix data download path

Rename ´data/´ to ´m3_data/´ to avoid user directory conflicts

v0.1.4: Merge pull request #26 from rafiattrach/fix-path-handling

Read-only filesystem errors when using pip install m3-mcp with Claude Desktop

Changed:
✅ Use home directory for data storage (pip installs)
✅ Remove noisy logs on m3 -v or m3 config
✅ Lazy directory creation

Install: pip install m3-mcp==0.1.4

Version 0.1.3 - Config Scripts Packaging Fix

🔧 Version 0.1.3 - Config Scripts Packaging Fix

Critical Bug Fix: Resolves m3 config command failing after pip install with "config script not found" error.

✅ Fixed

• Config scripts packaging: Move mcp_client_configs to proper location for pip distribution
• Data directory detection: Smart detection between development and production environments
• Module imports: Use proper module resolution instead of file paths

📝 Improved

• Version management: Single source of truth in src/m3/__init__.py

🧪 Verified

• ✅ Development workflow: git clone → pip install -e . → all commands working
• ✅ All 51 unit tests passing

v0.1.2 - Fix Package Distribution

🐛 Bug Fix

 - Fixed missing mcp_client_configs in package distribution
 - Resolves `m3 config` command failure after pip install
 - Package now includes all necessary configuration scripts

📦 Installation

pip install m3-mcp==0.1.2

v0.1.1 - Automated PyPI Publishing

Release description:

🚀 What's New

• Automated PyPI Publishing: Releases now automatically publish to PyPI
• Secure OIDC Authentication: No API keys needed - uses GitHub's trusted publishing
• Proper Team Attribution: All authors credited
• Tag-Based Versioning: Git tags automatically control PyPI versions

🔧 Technical Changes

• Added GitHub Actions workflow for secure PyPI publishing
• Environment protection with approval gates
• Comprehensive testing before each publish

📦 Installation

pip install m3-mcp

For contributors: Normal development workflow unchanged - just create GitHub releases when ready to publish!