Be the first to get access to new connectors. Get early access →
ConnectorBay Early access

MLflow on Databricks (BYOK) — connector documentation

  • MCP endpoint: https://byok-mlflow-databricks.connectorbay.com/mcp
  • REST API: POST https://byok-mlflow-databricks.connectorbay.com/v1/tools/{tool_name}, described by the OpenAPI 3.1 spec at https://byok-mlflow-databricks.connectorbay.com/openapi.json
  • Transport: Streamable HTTP, stateless, MCP specification 2025-06-18 or later
  • Authentication: your own MLflow on Databricks Databricks personal access token, sent in the X-Upstream-Key header of every request (never stored, logged or cached)
  • Health: https://byok-mlflow-databricks.connectorbay.com/health · Tool schemas: https://byok-mlflow-databricks.connectorbay.com/tools.json
  • Publisher: Unify Dynamics LLC · connectorbay.com/byok-mlflow-databricks · agent@connectorbay.com

What MLflow on Databricks (BYOK) is for

MLflow on Databricks (BYOK) lets an AI assistant read your own MLflow on Databricks data through the MLflow on Databricks API with a key you supply (Databricks personal access token). Read-only: 11 tools (search_experiments, get_experiment, get_experiment_by_name, search_runs, get_run, get_metric_history, list_run_artifacts, search_registered_models, get_registered_model, search_model_versions, get_model_version). The BYOK key travels with each request, is used for that one call only and is never stored, logged or cached; it is only ever sent to *.cloud.databricks.com. Unofficial connector by Unify Dynamics LLC: not affiliated with, endorsed by or sponsored by MLflow on Databricks. MLflow on Databricks and its logos are trademarks of their owners and are used only to say which service the connector works with.

Ask your assistant things like:

  • "Which MLflow experiments does my Databricks workspace have?"
  • "What is MLflow experiment 1234567890 in my workspace?"
  • "Find the MLflow experiment called /Shared/churn-model"
  • "Which runs in experiment 1234567890 have rmse below 1?"
  • "What were the metrics and parameters of MLflow run abc?"
  • "Show the loss curve of MLflow run abc"

What it does

  • search_experiments: Search MLflow experiments of the workspace by name or tag filter, with lifecycle stage, location and update time (Unix milliseconds).
  • get_experiment: Get one MLflow experiment by id: name, artifact location, lifecycle stage, tags and times.
  • get_experiment_by_name: Find one MLflow experiment by its exact name (on Databricks this is the workspace path such as /Users/me@example.com/churn).
  • search_runs: Search the runs of one experiment, optionally with a filter on metrics, params or tags, newest first, with their metrics and params.
  • get_run: Get one MLflow run: status, start and end time (Unix milliseconds), latest metrics, params and tags.
  • get_metric_history: List all logged values of one metric of a run with step and timestamp, for example a training loss curve.
  • list_run_artifacts: List the artifact files and folders of a run (one directory level) with path, size and whether it is a directory.
  • search_registered_models: Search the workspace model registry (legacy, non Unity Catalog) by name, with latest versions and update time.
  • get_registered_model: Get one registered model of the workspace registry: description, tags, latest versions and times.
  • search_model_versions: Search model versions of the workspace registry with a filter such as name='churn-model', with stage, status and source run.
  • get_model_version: Get one version of a registered model: stage, status, source run, description, tags and times.

BYOK key

You bring your own MLflow on Databricks Databricks personal access token (https://docs.databricks.com/aws/en/dev-tools/auth/pat). Your assistant sends it in the X-Upstream-Key header with each request. ConnectorBay uses it for that one call only: it is never stored, logged or cached, never shown in an answer, and it is only sent to *.cloud.databricks.com over https, with no redirects followed. Create a key with the fewest permissions the tools need (read-only if the provider offers it), and revoke it at the provider whenever you like.

Data sources

Live from the MLflow on Databricks API (*.cloud.databricks.com) with a BYOK key; documentation: https://docs.databricks.com/api/workspace/experiments. Endpoints checked against that documentation on 2026-10-16.

What it is not

Unofficial connector by Unify Dynamics LLC: not affiliated with, endorsed by or sponsored by MLflow on Databricks. MLflow on Databricks and its logos are trademarks of their owners and are used only to say which service the connector works with. It is read-only: it never creates, changes, sends or deletes anything in your MLflow on Databricks account.

Built for: Worldwide. Languages: English.