📖 Databricks Apps Cookbook 🍳

Ready-to-use code snippets for building Dash, Streamlit, Reflex, and FastAPI apps on Databricks Apps.

Learn more on apps-cookbook.dev. Coding agents: start with AGENTS.md and the path below.

What is the Databricks Apps Cookbook?

• 10+ recipes for common Apps use cases such as reading and writing tables and volumes, invoking ML / GenAI, or triggering workflows.
• Try recipes in the Cookbook app and copy a snippet into your own app.
• Requirements (permissions, resources, dependencies) for each recipe.
• Deploy to Databricks Apps or run locally.
• Snippets for Dash, Streamlit, Reflex, and FastAPI only.

Choose a path

This repo is the Python recipe layer for Dash, Streamlit, Reflex, and FastAPI.

MCP

Coding agents (Cursor, Claude Code, and others)

Cookbook skills teach Dash/Streamlit/Reflex/FastAPI composition. The recipe MCP (list_cookbook_recipes, get_cookbook_recipe) is how an agent in an empty app folder gets snippets without guessing. Official platform skills (AppKit, apps deploy) still come from AI Dev Kit / databricks aitools. Human walkthrough (Cursor, Claude, upgrade): Use with coding agents on the docs site.

You only need git, bash, and Python 3. Clone from GitHub; skills and the recipe MCP install from this repo (public PyPI for the MCP venv). You do not need a Databricks VPN or internal npm. Node.js 20+ is only if you preview the docs site.

Install — project vs global

Same idea as AI Dev Kit: project (default) is the folder you run from; --global is this user on this machine.

From this cookbook clone:

# This repo only
./install.sh
./mcp-server/mcp_install.sh

# Every repo on this machine
./install.sh --global
./mcp-server/mcp_install.sh --global

Cursor MCP has no global config file (same as the kit). Use project .cursor/mcp.json or Cursor Settings → MCP. After install, enable the cookbook server if it is off.

New app folder (keep the cookbook clone on disk — MCP runs from there):

COOKBOOK=/path/to/databricks-apps-cookbook   # this clone
mkdir -p ~/tmp/cookbook-scratch && cd ~/tmp/cookbook-scratch && git init
"$COOKBOOK/install.sh" --target-dir "$PWD"
"$COOKBOOK/mcp-server/mcp_install.sh" --target-dir "$PWD"

That writes the same skill and MCP files AI Dev Kit uses, including .claude/skills/, .cursor/skills/, .mcp.json, and .cursor/mcp.json. Default --tools is Claude, Cursor, Copilot, Codex, Gemini, Antigravity, Windsurf, OpenCode, and Kiro. Narrow it with --tools claude,cursor.

Claude Code can load skills as a plugin (no copied folders). In Claude:

/plugin marketplace add databricks-solutions/databricks-apps-cookbook
/plugin install databricks-skills@databricks-skills

How to pick up the next repo release: Upgrade. Full plugin commands: databricks-skills/README.md.

Try it in Cursor

1. File → Open Folder on the app (for a from-scratch test: ~/tmp/cookbook-scratch, not this cookbook).

2. Settings → MCP: turn on cookbook. Restart Cursor if the server is missing.

3. Prompt that should use cookbook recipes:

   Build a Streamlit Databricks App that reads a Unity Catalog table. Use Databricks Apps Cookbook recipes only. Do not invent code.

   You should see the build-app and tables skills, cookbook MCP (list_cookbook_recipes / get_cookbook_recipe), then app.yaml and Streamlit table-read code from this repo.

4. Prompt that should not scaffold from this cookbook:

   I want a TypeScript/React Databricks App.

   The assistant should point at AppKit / AI Dev Kit (databricks apps init) instead of copying Dash or Streamlit from here.

Try it in Claude Code

cd ~/tmp/cookbook-scratch && claude

Confirm a cookbook MCP server, then use the same two prompts as Cursor.

Build a Python app with an agent

1. Clone or open this repository, or a target app with --target-dir as above, so the agent can load skills (and MCP, if registered).
2. Tell the agent the framework: Dash, Streamlit, Reflex, or FastAPI.
3. The agent should load databricks-skills/build-app/SKILL.md, then the matching category skill (tables, authentication, …).
4. Implementations must come from the recipe index → docs/docs//… plus the sample modules (or MCP get_cookbook_recipe). Do not invent AppKit, Gradio, or Flask (no tested recipes here).
5. When the app runs, optionally follow databricks-skills/productionize-app-dab/SKILL.md for Databricks Asset Bundles.

Client path tables: databricks-skills/README.md, mcp-server/README.md. Isolated installer test: ./mcp-server/tests/test_install_paths.sh.

Upgrade

Official Databricks skills (AppKit, jobs, …) are not this repo: databricks aitools update.

This cookbook versions with git (and GitHub Releases). Copied skills do not auto-update; the recipe MCP reads whatever is in this clone.

Subscribe to releases

1. Open databricks-solutions/databricks-apps-cookbook.
2. Watch → Custom → enable Releases (and Releases only if you do not want all issues/PRs).
3. Release notes: Releases.

On the next repo release

Use the same flags as the first install (--global, --target-dir, --tools).

cd /path/to/databricks-apps-cookbook
git fetch --tags
git pull                         # or: git checkout 
./install.sh                     # overwrites skill copies in place
./mcp-server/mcp_install.sh      # refreshes mcp-server/.venv; catalog is the pulled docs

Claude Code plugin (no skill copies):

/plugin marketplace update databricks-skills
/reload-plugins

Restart Cursor / Claude Code. MCP list_cookbook_recipes then sees new .mdx recipes from the pull; you do not re-register MCP unless the installer or venv changed.

Uninstall only cookbook entries:

./install.sh --uninstall
./mcp-server/mcp_install.sh --uninstall
# add --global if that is how you installed

Documentation

Find deployment instructions and all code snippets on apps-cookbook.dev.

Recipe index by framework

Where things live

• Recipe write-ups: .mdx files under docs/docs//…, published at apps-cookbook.dev.
• Agent skills: databricks-skills/README.md — each SKILL.md is under databricks-skills// (for example databricks-skills/build-app, databricks-skills/productionize-app-dab), not under framework folders.
• Runnable sample apps: dash/APP_DESCRIPTION.md, streamlit/APP_DESCRIPTION.md, reflex/APP_DESCRIPTION.md, fastapi/APP_DESCRIPTION.md. Each describes this app’s pages and links back to the recipe index, docs site, contributing guide, and deploy instructions. FastAPI also has fastapi/README.md for run commands and endpoints.

How to read Doc path

Every cell is relative to docs/docs//. On disk, add .mdx. Example: Streamlit “Read Delta table” is docs/docs/streamlit/tables/tables_read.mdx (site: /docs/streamlit/tables/tables_read).

Most recipes use that same relative path in every framework that has a checkmark. A few do not; those cells list each path and which frameworks it applies to. Today that is:

• Retrieve secrets — Dash: external_services/secrets_retrieve. Streamlit and Reflex: authentication/secrets_retrieve.
• OLTP / Postgres — Dash: tables/oltp_database. Reflex: tables/oltp_database_connect.

FastAPI endpoint recipes usually live under building_endpoints/ instead of tables/ or aiml/.

Shared recipes (Dash, Streamlit, Reflex, FastAPI)

Sample code: dash/pages/ · streamlit/views/ · reflex/app/pages/ · fastapi/routes/. Reflex page modules are listed in reflex/APP_DESCRIPTION.md.

FastAPI-only guides

All under docs/docs/fastapi/:

Contributions

We welcome contributions! See CONTRIBUTING.md — a recipe is sample + docs + the index below; it does not become a skill. Add or change a category SKILL.md only when agents need new composition notes (or a new category). Submit a pull request or raise an issue. After a release, users upgrade from GitHub Releases.

Not sure what to contribute? Here are some commonly requested samples:

• Write data from a form into a Delta table
• Display coordinates from a Delta table in a map component
• Display data from a Delta table in Streamlit/Dash-native diagram components
• Gradio implementation
• Flask implementation

Support

These samples are experimental and meant for demonstration purposes only. They are provided as-is and without formal support by Databricks. Ensure your organization’s security, compliance, and operational best practices are applied before deploying them to production.

License

© 2025 Databricks, Inc. All rights reserved. The source in this notebook is provided subject to the Databricks License. All included or referenced third party libraries are subject to the licenses set forth below.