← Projects · MCP Protocol · MLOps · PyPI · Open Source

MLOps MCP Server

FastMCP · MLflow · DVC · Git · SciPy · Python 3.11+

An open-source MCP server that gives AI assistants tool-based access to MLOps workflows: experiment tracking, model registry, dataset management, pipeline DAG validation and lineage tracing. Built and published by me; installable by anyone.

pip install mlops-mcp-server   # v0.1.0 · MIT · Python 3.11+

Problem

AI coding assistants can write training code but cannot see the MLOps state around it: which run is best, which model version is current, whether a dataset has drifted, or whether a pipeline DAG is valid. This server lets the assistant call typed tools and work on the ML workflow directly.

My role

Sole author: designed, built, documented and published the package.

What is inside

15 domain modules covering file operations, analysis, project scaffolding, archives, cleanup, experiments, datasets, model registry, Git, DVC, MLflow, pipelines, environments, lineage and documentation, with a small always-available core that includes module discovery and activation tools.

Tech stack

Key design decisions & trade-offs

• Module activation API versus startup registration. The server includes discovery and activation tools designed to load domain modules on demand and keep the assistant’s context small. In v0.1.0, all tools are registered at startup, because not every MCP host refreshes its tool list during a session. Gating registration by host capability is the next step.
• Real algorithms behind the tools. Dataset drift detection uses the two-sample Kolmogorov-Smirnov test; pipeline validation uses Kahn’s algorithm for cycle detection; lineage tracing is breadth-first search over the provenance graph, rendered as Mermaid diagrams.
• Optional dependencies. Heavy backends such as MLflow are optional extras, so the base install stays light and features degrade gracefully when a backend is missing.
• Published, versioned, licensed. v0.1.0 on PyPI under the MIT licence.

Results

Published on PyPI (v0.1.0, MIT) · pip-installable by anyone · 15 domain modules · integrates MLflow, DVC and Git