Comet MCP Server
Comet ML experiment tracking data available directly within your AI tools.
Data last scanned today · Reviewed today
Overview
Comet MCP Server provides a bridge between your AI development environment and the Comet ML experiment tracking platform. It enables users to programmatically list projects, search for specific experiments, and retrieve detailed performance metrics, source code, and training logs without leaving their chat interface. The server supports structured interactions with your Comet ML workspace, including tools to count project experiments and fetch session metadata. It handles authentication through standard Comet ML configuration methods, such as environment variables, ensuring consistent access to your experiment history.
Our verdict
This server carries a high risk level with a security score of 29/100. Key contributors to this assessment include its status as community-maintained rather than official and the lack of an editorially reviewed authentication method. Given these factors and the limited community presence indicated by a single star and low contributor count, it is best suited for experimental use in non-sensitive environments where strict audit requirements are not a priority.
- Supports zero direct dependencies to minimize supply-chain risks.
- Provides documented tools for deep experiment and project inspection.
- Includes robust error handling for API interactions.
- Allows deployment via Docker for containerized isolation.
- Security score of 29/100 due to community maintenance status.
- Authentication methods remain unreviewed and unverified.
- Limited community visibility with only one GitHub star.
Tools
| Tool | Description | Risk |
|---|---|---|
| list_experiments | List recent experiments with optional filtering | low |
| get_experiment_details | Get comprehensive experiment information including metrics and parameters | low |
| get_experiment_code | Retrieve source code from experiments | low |
| get_experiment_output | Get raw training logs (stdout/stderr) from experiments | low |
| get_experiment_metric_data | Get metric data for multiple experiments | low |
| get_default_workspace | Get the default workspace name for the current user | low |
| list_projects | List all projects in a workspace | low |
| list_project_experiments | List experiments within a specific project | low |
| count_project_experiments | Count and analyze experiments in a project | low |
| get_session_info | Get current session status and connection information | low |
Compatibility
| Client | Local | Docker | Remote | Read-only |
|---|---|---|---|---|
| Claude Desktop | ||||
| Cursor | ||||
| VS Code | ||||
| Windsurf | ||||
| ChatGPT |
Frequently asked questions
›What are the prerequisites for running the Comet MCP server?
You need Python 3.8 or higher and a valid Comet ML account with an API key.
›How do I configure my credentials for the server?
You can configure your credentials by running the comet init command, or by setting the COMET_API_KEY and COMET_WORKSPACE environment variables.
›Can I run the server without installing Python dependencies locally?
Yes, you can build and run the provided Docker image to avoid installing Python dependencies on your local system.
›How do I access files like CSV exports generated by the tools?
Tools that generate files return a resource URI; you can access the content by using the MCP read_resource method with that URI.
›Which transport modes does the server support?
The server supports both stdio and sse transport modes, which can be configured via command line options.
Alternatives
Changelog
Badge
Maintain this server? Add the live badge to your README.