Installation

flowx is a set of agent skills that run from an AI coding assistant. Skills can be run from:

Databricks Genie Code using an MCP server hosted in Databricks Apps
Local agent harnesses (e.g. Claude Code) using either a Python virtual environment or a local MCP server

The following sections provide instructions for installing flowx using your preferred coding assistant.

Installing flowx for Databricks Genie Code

flowx can be used directly from Databricks Genie Code. To install flowx for Genie Code, you must deploy the flowx MCP server as a Databricks App in your workspace, then register flowx as a custom MCP server.

Clone flowx into a shared workspace folder

First clone the repo into your workspace using a Git folder.

Use a shared folder

The flowx MCP app uses a service principal that cannot read private folders in /Workspace/Users/. Clone flowx into /Workspace/Shared to allow source code access for all workspace users and service principals. If your workspace restricts access to /Workspace/Shared, use an approved shared folder and configure the deployment notebook to deploy the flowx MCP app from this shared folder.

Copy the skills into your assistant folder

Genie Code references skills in your .assistant/skills folder automatically. These can either be installed for a single user or for all users within a workspace.

To add flowx skills for a single user, copy the skills from the flowx folder to your .assistant folder:

bash

databricks workspace import-dir skills /Users/<EMAIL_ADDRESS>/.assistant/skills

To add flowx for all workspace users, add skills to a workspace-level folder:

bash

databricks workspace <FLOWX_DIRECTORY> skills /Workspace/.assistant/skills

Deploy the MCP app

Open and run app/deploy_app.py. Set the repo_root widget to the location of your flowx folder (e.g. /Workspace/Shared/flowx). flowx uses the Databricks SDK to stage a Databricks Declarative Automation Bundle to /Workspace/Shared/mcp-flowx and deploy the app. The notebook prints the app URL when deployment succeeds. The MCP endpoint is <APP URL>/mcp.

Installing the flowx MCP using the Databricks CLI

You can also run bash app/deploy.sh from a shell terminal to deploy the flowx MCP app. This requires a Databricks CLI session and cannot be run from serverless compute.

Grant access to the MCP app

Grant CAN USE access on the mcp-flowx app to any users or service principals that will use flowx. Grant the app's service principal READ VOLUME access to any workspace folders or Unity Catalog volumes where your pipeline files are stored.

Register the MCP server in Genie Code

MCP servers are available in Genie Code Agent mode:

In the Genie Code panel, click Settings.
Under MCP Servers, click Add Server.
Choose Custom MCP server and select the mcp-flowx Databricks App.
Click Save.

This installs the flowx tool which is called when you run migrations using flowx.

Genie Code custom MCP requirements

A custom MCP app must be deployed in the same workspace and reachable at https://<APP URL>/mcp. If Genie Code cannot connect, set the app's FLOWX_ALLOWED_ORIGINS environment variable to your workspace URL and redeploy. See Connect Genie Code to MCP servers.

Verify

Ask Genie Code "What flowx MCP tools are available?". You should see the single flowx tool. You can now run @flowx-migrate and other flowx skills.

Troubleshooting the flowx MCP app

If Genie Code does not have access to the flowx MCP tool, open your app's health endpoint at <APP URL>/. This should return {"status":"ok"}). If you cannot access the health endpoint, verify the deployment steps and redeploy the app.

Installing flowx for Claude Code

flowx can also be installed with local agent harnesses (e.g. Claude Code) as an agent skills plugin. Skills can run commands from a Python virtual environment or locally-hosted MCP server.

Install the plugin

flowx is distributed through a Claude Code marketplace. It can be installed as a plugin or directly copied into your skills folder.

bash

/plugin marketplace add databricks-solutions/flowx
/plugin install flowx@flowx
/reload-plugins

You can also copy the skill folders straight into your local skills directory:

shell

cp -R skills/{flowx-setup,flowx-discover,flowx-convert,flowx-package,flowx-migrate} ~/.claude/skills/

Run the setup skill

Run /flowx:flowx-setup to set up a local Python environment with flowx's required 3rd-party packages. The setup process will:

Check that python3, pip, and the venv module are available.
Create the virtual environment at <plugin_dir>/.venv.
Install packages listed in requirements.txt into the virtual environment with pip.
Write the interpreter path to a marker file (<plugin_dir>/.migration-venv) used when running the flowx skills.

Virtual environment creation and reuse

The environment is created once and reused. No uv is required for plugin users.

Verify

Open Claude Code and ask "What flowx skills do you have available?". You should see a list of skills (e.g. flowx-setup, flowx-migrate). You can now run /flowx:flowx-migrate, /flowx:flowx-discover, and other flowx skills.

If you hit a ModuleNotFoundError while running a phase, the venv is missing or incomplete — re-run /flowx:flowx-setup. Every Python command the skills run uses the interpreter recorded in <plugin_dir>/.migration-venv, with src/ on PYTHONPATH:

bash

export PYTHONPATH="<plugin_dir>/src"
PY="$(cat <plugin_dir>/.migration-venv)"
"$PY" -m flowx.adapter inputs discover --source adf   # or --source airflow

Other Agent tools

Any tool that supports the Agent Skills open standard can install flowx:

Copy each skill folder (e.g. skills/flowx-migrate) into the tool's configured skills folder. Ensure the folder contains SKILL.md directly.
Restart the tool if it caches skill metadata at startup.
Run the flowx-setup skill to create flowx's Python environment

Flattening skills files

If your tool expects a single Markdown file instead of a directory tree, run the following command to concatenate the skills:

bash

cat skills/*/SKILL.md > flowx-skills.md

Installation

On this page