mlflow-mcp 0.1.0

A Model Context Protocol (MCP) server for MLflow - enables LLMs to interact with MLflow experiments, runs, metrics, and models

MLflow MCP Server

A Model Context Protocol (MCP) server that enables LLMs to interact with MLflow tracking servers. Query experiments, analyze runs, compare metrics, and explore the model registry - all through natural language.

Features

• Experiment Management: List and search experiments, discover available metrics and parameters
• Run Analysis: Retrieve run details, query runs with filters, find best performing models
• Metrics & Parameters: Get metric histories, compare parameters across runs
• Artifacts: Browse and download run artifacts
• Model Registry: Access registered models, versions, and deployment stages
• Comparison Tools: Side-by-side run comparisons, best run selection
• Tag-based Search: Filter runs by custom tags

Installation

Using uvx (Recommended)

# Run directly without installation
uvx mlflow-mcp

# Or install globally
pip install mlflow-mcp

From Source

git clone https://github.com/kirillkruglikov/mlflow-mcp.git
cd mlflow-mcp
uv sync
uv run mlflow-mcp

Configuration

Claude Desktop

Add to your Claude Desktop config file:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json Windows: %APPDATA%\Claude\claude_desktop_config.json Linux: ~/.config/claude/claude_desktop_config.json

{
  "mcpServers": {
    "mlflow": {
      "command": "uvx",
      "args": ["mlflow-mcp"],
      "env": {
        "MLFLOW_TRACKING_URI": "http://localhost:5000"
      }
    }
  }
}

Environment Variables

• MLFLOW_TRACKING_URI (required): Your MLflow tracking server URL
  • Examples: http://localhost:5000, https://mlflow.company.com

Available Tools

Experiments

• get_experiments() - List all experiments
• get_experiment_by_name(name) - Get experiment by name
• get_experiment_metrics(experiment_id) - Discover all unique metrics
• get_experiment_params(experiment_id) - Discover all unique parameters

Runs

• get_runs(experiment_id, limit=10) - Get runs for an experiment
• get_run(run_id) - Get detailed run information
• query_runs(experiment_id, query, limit=10) - Filter runs (e.g., "metrics.accuracy > 0.9")
• search_runs_by_tags(experiment_id, tags) - Find runs by tags

Metrics & Parameters

• get_run_metrics(run_id) - Get all metrics for a run
• get_run_metric(run_id, metric_name) - Get full metric history with steps

Artifacts

• get_run_artifacts(run_id, path="") - List artifacts (supports browsing directories)
• get_run_artifact(run_id, artifact_path) - Download artifact
• get_artifact_content(run_id, artifact_path) - Read artifact content (text/json)

Analysis & Comparison

• get_best_run(experiment_id, metric, ascending=False) - Find best run by metric
• compare_runs(experiment_id, run_ids) - Side-by-side comparison

Model Registry

• get_registered_models() - List all registered models
• get_model_versions(model_name) - Get all versions of a model
• get_model_version(model_name, version) - Get version details with metrics

Health

• health() - Check server connectivity

Usage Examples

Ask Claude

"Show me all experiments in MLflow"

"What are the top 5 runs by accuracy in experiment 'my-experiment'?"

"Compare runs abc123 and def456"

"Which model has the highest F1 score?"

"Show me the training loss curve for run xyz789"

"List all production models in the registry"

Development

# Install dependencies
uv sync

# Run server
uv run mlflow-mcp

# Run tests
uv run pytest

# Format code
uv run black src/
uv run ruff check src/

Requirements

• Python >=3.10
• MLflow >=3.4.0
• Access to an MLflow tracking server

License

MIT License - see LICENSE file for details.

Contributing

Contributions welcome! Please open an issue or submit a pull request.

Links

• GitHub Repository
• MLflow Documentation
• Model Context Protocol