ClearML

You can run a lightweight MCP server to connect AI assistants with ClearML experiments, models, and projects. It gives you full context, metrics, and artifacts directly in conversations, enabling practical analysis and discovery across your ClearML setup.

How to use

You interact with the ClearML MCP server from your MCP-enabled assistant. Start the server locally using one of the available runtime commands, then point your assistant to that local server. Once running, you can ask questions like to find experiments, compare metrics, inspect hyperparameters, and retrieve artifacts. The MCP server exposes a set of tools that let you explore tasks, models, projects, and analysis results, all through natural language prompts.

How to install

Prerequisites: make sure you have Python installed and that you can run the MCP runtime through UVX or Python. You also need a configured ClearML environment with credentials in your ClearML config file.

# Install the MCP server package from PyPI
pip install clearml-mcp

# Or run directly with UVX (no installation needed)
uvx clearml-mcp

Additional setup notes

The MCP server uses your existing ClearML credentials stored in ~/.clearml/clearml.conf. You should ensure these credentials are valid and have permission to access the projects, tasks, and artifacts you plan to query.

If you prefer to run the MCP server without installing, you can invoke the runtime directly via Python in a development setup.

Development and quick-start commands

# Install the MCP server package from PyPI
pip install clearml-mcp

# Run the MCP server using UVX (no npm install required)
uvx clearml-mcp

# Alternative: run via Python module (requires a Python install)
python -m clearml_mcp.clearml_mcp

What you can do with the MCP server

With the server running, you can ask your AI assistant to perform tasks such as discovering experiments, comparing metrics across tasks, inspecting hyperparameters, retrieving artifacts, and listing projects. This enables practical ML workflow insights directly inside your conversations.

Troubleshooting tips

If you encounter connection or performance issues, verify that your ClearML credentials are correct in ~/.clearml/clearml.conf and that your ClearML server is reachable from your environment. If you have trouble starting the server, try the Python module path or the UVX command as shown above.

Available tools

get_task_info

Fetch detailed information about a specific ClearML task, including parameters, status, and metadata.

list_tasks

List tasks with advanced filtering by project, status, tags, and user.

get_task_parameters

Retrieve hyperparameters and configuration for a task.

get_task_metrics

Access training metrics, scalars, and plots for tasks.

get_task_artifacts

Retrieve artifacts, model files, and outputs associated with a task.

get_model_info

Get metadata and configuration details for a model.

list_models

Browse available models with filtering options.

get_model_artifacts

Access model files and download URLs.

list_projects

Discover available ClearML projects.

get_project_stats

Get project statistics and task summaries.

find_project_by_pattern

Find projects matching name patterns.

find_experiment_in_project

Find specific experiments within projects.

compare_tasks

Compare multiple tasks by specific metrics.

search_tasks

Perform advanced search by name, tags, comments, and more.