Summary: Build and publish V1 of the MongoDB storage backend for MLflow (PyPI)

Background

MLflow’s storage layer is pluggable but ships only with file-based and SQL backends. This ticket covers building a MongoDB-backed Tracking and Model Registry store plugin for MLflow, and publishing it to PyPI so users can install it as a drop-in plugin. A customer wants to use it immediately, so speed matters.

Scope

Full V1 scope, requirements, and fast-follow items: V1 Technical Scope

Work

1. Build
   • Implement the MongoDB store plugin in Python, starting with the Prompts MVP and then the rest of V1.
   • Support MLflow 3.5+, Python 3.11+, MongoDB 8.0+.
   • Add unit and integration tests.
2. Publish
   • Package it as an installable plugin (pyproject, entry points, versioning).
   • Set up the release process and publish to PyPI.
   • Add a README and quickstart.

Delivery

V1 of the MongoDB backend for MLflow, published on PyPI.

Acceptance Criteria

• The V1 scope in the linked doc is implemented, and the Prompts MVP is delivered first.
• Tests pass on the supported MLflow, Python, and MongoDB versions.
• The package is published on PyPI, and pip install plus MLflow configuration works as documented.
• The README and quickstart are included.