Summary
The CML loop — edit, run, check — without the web UI.
It covers the whole cycle a data scientist repeats all day: change a file in the project, turn it into a job, run it, read the run. Because file patching is a tool of its own, an assistant can fix and re-run without a full upload each time.
What it is
A standalone MCP server for Cloudera Machine Learning, covering projects, project files, jobs and runs against a CML instance you name.
What you get
- Projects listed, created and inspected without opening the CML web interface
- Project files read, uploaded, renamed and patched in place — targeted edits rather than re-uploading a whole file
- Jobs created from scratch or straight from a file in the project, then listed and inspected
- Job execution controlled end to end: run now, schedule, list previous runs, stop a run that is going wrong
- The available runtime addons listed, so a job is configured against what the instance actually offers
- SSL certificate download for instances using a self-signed certificate
Requirements
Python 3.8+ with the MCP CLI package and requests. The instance is addressed by its host URL and an API token, given as environment variables — the server accepts more than one naming scheme — or as the same values on the launch command. Self-signed instances need the certificate downloaded and the server pointed at that file.