Overview
The Optuna MCP Server is a Model Context Protocol (MCP) server that automates optimization and analysis tasks by leveraging Optuna. It exposes a set of primitive tools from Optuna—Study, Trial, Visualization, and a Web Dashboard—so MCP clients can orchestrate hyperparameter optimization without memorizing internal APIs. The server supports creating and loading studies, configuring samplers, managing metric names, retrieving directions and trials, and exporting results in CSV. Through Trial, clients can propose parameters (ask) and report results (tell), while also setting and querying trial attributes. Visualization tools enable plots of optimization history, hypervolume history, Pareto fronts, contours, parallel coordinates, and more, including EDF, timelines, and rank visuals. The Web Dashboard can be launched to analyze results interactively, defaulting to port 58080. Installations can be done via uv or Docker, with optional storage backends like SQLite. It is designed to support automated workflows from large language models, interactive analysis via chat interfaces, and integration with other MCP tools.
Features
Study
Create and load Optuna studies, configure directions, and manage study-related metadata (e.g., sampler, metrics) via MCP.
Trial
Propose new trial parameters (ask), report results (tell), and manage trial attributes and metadata.
Visualization
Provide plotting tools for optimization history, hypervolume, Pareto fronts, contours, parallel coordinates, and related visual analyses.
Web Dashboard
Launch the Optuna dashboard for interactive analysis of optimization results (default port 58080).
Who Is This For?
- LLM users:Automate hyperparameter optimization workflows using LLM prompts and responses within an MCP-enabled environment.
- Data scientists:Interactively analyze Optuna optimization results through a chat interface integrated with MCP.
- MCP integrators:Assist other MCP tools by optimizing their input and output flows through the MCP server.
