clearml

The ClearML MCP Server provides a dedicated interface to query and analyze ClearML experiments, models, projects, and related artifacts. It exposes a suite of 14 tools that let you fetch task information, parameters, metrics, artifacts, and model details, as well as perform cross-task comparisons and advanced searches. This makes it easy for your AI assistant or editor to surface experiment data, inspect training progress, compare performance across runs, and retrieve relevant artifacts or model files directly during conversations or workflows.

To use it, install the MCP server (see installation steps), then run it via the supported runtime (for example UV with Python). Once running, configure your MCP-enabled assistant or editor to point at the clearml-mcp server. Typical configurations use the UV-based command with the server name clearml, e.g., uvx clearml-mcp. From there you can issue natural-language prompts like: "Show the latest experiments in the computer-vision project" or "Compare the accuracy metrics between tasks task-123 and task-456". The server will interpret these requests and return structured results from ClearML, including task details, metrics, parameters, and artifacts, enabling seamless integration into your AI-assisted workflows.

Prerequisites:

• Python 3.10+ and access to install Python packages
• ClearML credentials configured in ~/.clearml/clearml.conf
• Optional: uv (uvx) for running without installation

Installation steps:

1. Install the MCP server from PyPI:

pip install clearml-mcp

2. Run directly with uvx (no installation required):

uvx clearml-mcp

3. Alternatively, run as a Python module (if you prefer direct Python invocation):

uvx clearml-mcp
# or explicitly
python -m clearml_mcp.clearml_mcp

4. Verify installation by listing or querying a simple task (configured ClearML credentials required). See the troubleshooting section if you encounter connection errors.

Tips and considerations:

• Ensure your ClearML credentials are present in ~/.clearml/clearml.conf to enable API access.
• If you hit connectivity or project access issues, validate credentials with the ClearML Python API, e.g., python -c "from clearml import Task; print(Task.get_projects())".
• The MCP server supports Python-based runtimes and UV-based workflows; you can integrate with Claude, Cursor, Continue, Cody, and other MCP-enabled assistants by configuring the appropriate mcpServers entry (see integrations in the README).
• When querying large datasets, prefer filtering by project, status, or tags to improve performance.
• For testing MCP inspector, you can simulate a client query with: npx @modelcontextprotocol/inspector uvx clearml-mcp