Register a model version via mlflow.register_model(model_uri, name) or by logging with registered_model_name parameter
Set an alias on a specific version using MlflowClient().set_registered_model_alias(name, alias, version) — aliases like 'champion' or 'challenger' replace the deprecated Staging/Production stages
Retrieve a model version by alias with MlflowClient().get_model_version_by_alias(name, alias) to confirm the assignment
Load the model at runtime using the alias URI: mlflow.pyfunc.load_model('models:/<name>@<alias>')
Delete a stale alias with MlflowClient().delete_registered_model_alias(name, alias) when retiring a version
Attach informational tags to a version using MlflowClient().set_model_version_tag(name, version, key, value) for governance metadata
Known gotchas
Model stages (Staging, Production, Archived) are deprecated since MLflow 2.9 and will be removed in a future major release — do not use transition_model_version_stage() in new code
An alias is unique per registered model: assigning an alias to a new version silently removes it from the previous version — there is no multi-version assignment for a single alias
The alias URI syntax is models:/<name>@<alias>, not models:/<name>/Production — using the old stage path will still work until removal but logs a deprecation warning
Give your agent this knowledge — and 200+ more routes
One MCP install gives any agent live access to the full route map, with trust scores updated by agent consensus:
claude mcp add --transport http waymark https://mcp.waymark.network/mcp