Domino Data Lab Plugin for Claude Code, ChatGPT, and Codex

One plugin, two distributions. This repository is a Claude Code plugin as checked in, and it also builds a ChatGPT / Codex plugin package for the OpenAI plugin directory. Both share the same skills and the same Domino MCP server.

Overview

The plugin helps AI assistants with all aspects of Domino Data Lab, including:

• Workspaces: Jupyter, VS Code, RStudio configuration and management
• Jobs: Batch execution, scheduled jobs, and monitoring
• Environments: Custom Docker environments and package management
• Datasets: Data versioning with snapshots and sharing
• NetApp Volumes: Enterprise-grade multi-terabyte storage with near-instant snapshots
• Apps: Deploy React, Streamlit, and Dash applications
• Models: Deploy, monitor, and manage model endpoints
• GenAI: Trace and evaluate AI agents with the Domino SDK
• Distributed Computing: Spark, Ray, and Dask clusters
• And more...

What ships where

| Component | Claude Code | ChatGPT / Codex | | --- | --- | --- | | skills/ (30 skills) | ✅ | ✅ | | Domino MCP server (mcp-servers/) | ✅ via .mcp.json (stdio) | Codex local install: ✅ via mcp.json (stdio). Public directory: needs a hosted HTTPS endpoint (see OpenAI build) | | Subagents (agents/) | ✅ | ❌ Claude-only | | Output styles (output-styles/) | ✅ | ❌ Claude-only | | Hook examples (hooks/) | Docs only | ❌ | | Manifest | .claude-plugin/plugin.json | plugin.json (Agent Plugins format) |

Skills are written provider-neutral. Where an instruction really is host-specific (for example which instructions file or MCP config file to write), the skill gives both the Claude Code and the Codex variant. CI enforces this; see Skill rules.

---

Installation: Claude Code

Prerequisites

• Claude Code CLI v1.0.33 or later (claude --version to check)
• Access to a Domino Data Lab instance
• Domino API key (for API operations when running outside a Domino workspace)
• uv package manager (install guide) — required for the bundled Domino MCP server

Option 1: Marketplace Install (Recommended)

This approach registers the plugin through Claude Code's native marketplace system so it persists across sessions.

Step 1: Clone the repository and create a marketplace wrapper

git clone https://github.com/dominodatalab/domino-ai-plugin.git

mkdir -p ~/.claude/marketplaces/domino/.claude-plugin
mkdir -p ~/.claude/marketplaces/domino/plugins

mv domino-ai-plugin ~/.claude/marketplaces/domino/plugins/domino-claude-plugin

Step 2: Create the marketplace manifest

cat > ~/.claude/marketplaces/domino/.claude-plugin/marketplace.json << 'EOF'
{
  "name": "domino-marketplace",
  "owner": {
    "name": "Domino Data Lab",
    "email": "support@dominodatalab.com"
  },
  "plugins": [
    {
      "name": "domino-claude-plugin",
      "description": "Domino Data Lab plugin - workspaces, jobs, environments, datasets, apps, models, and more",
      "source": "./plugins/domino-claude-plugin",
      "category": "development"
    }
  ]
}
EOF

Why domino-claude-plugin? Claude Code tracks a marketplace install as <entry name>@<marketplace>, so the entry keeps the name it had before the repo moved. Existing installs and Domino workspaces keep updating without a reinstall. Only the clone URL changed.

Step 3: Register the marketplace and install the plugin

Launch Claude Code and run:

/plugin marketplace add /home/<your-username>/.claude/marketplaces/domino
/plugin install domino-claude-plugin@domino-marketplace

Note: Replace <your-username> with your actual username, or use the full absolute path (e.g., /home/ubuntu/.claude/marketplaces/domino). The ~ shorthand may not expand correctly.

Step 4: Restart Claude Code, then verify with /plugin → Installed tab.

Option 2: Direct Plugin Directory (Development / Quick Start)

git clone https://github.com/dominodatalab/domino-ai-plugin.git
claude --plugin-dir ./domino-ai-plugin

Loaded this way the plugin is read in place: a git pull (or checking out a release-* branch or tag) takes effect at the next session with no reinstall.

Option 3: Team / Project-Level Install

For teams sharing a project, add the marketplace to your project's .claude/settings.json:

{
  "extraKnownMarketplaces": {
    "domino-marketplace": {
      "source": {
        "source": "directory",
        "path": "/path/to/domino-marketplace"
      }
    }
  },
  "enabledPlugins": {
    "domino-claude-plugin@domino-marketplace": true
  }
}

When team members trust the repository folder, Claude Code will prompt them to install the marketplace and plugin automatically.

Updating (Claude Code)

• Local marketplace install (Options 1 and 3): Claude Code reads the clone in place, so checking out a newer commit is the update; start a new session afterwards. git pull fails on a detached checkout, so fetch and check out main instead:

  git -C ~/.claude/marketplaces/domino/plugins/domino-claude-plugin fetch origin main
  git -C ~/.claude/marketplaces/domino/plugins/domino-claude-plugin checkout -B main origin/main
  

  If git -C ~/.claude/marketplaces/domino/plugins/domino-claude-plugin remote get-url origin still names domino-claude-plugin, run git -C <that path> remote set-url origin https://github.com/dominodatalab/domino-ai-plugin.git first. The two repositories have unrelated histories, so git pull alone can't switch a clone; the fetch and checkout -B above can.

• Anthropic marketplace: claude plugin update dominodatalab@claude-plugins-official, then restart. That entry follows whatever commit Anthropic pins, which still points at the deprecated domino-claude-plugin repository.

• --plugin-dir (Option 2): the plugin is read in place. git pull, then start a new session.

---

Installation: Codex and the ChatGPT desktop app

From this repository (local marketplace)

Codex reads the portable plugin.json, skills/, and mcp.json straight from the repo, so a local checkout works without a build step.

git clone https://github.com/dominodatalab/domino-ai-plugin.git ~/.codex/plugins/domino-ai-plugin

Then add an entry to ~/.agents/plugins/marketplace.json:

{
  "name": "domino-local",
  "interface": { "displayName": "Domino" },
  "plugins": [
    {
      "name": "dominodatalab",
      "source": { "source": "local", "path": "./.codex/plugins/domino-ai-plugin" },
      "policy": { "installation": "AVAILABLE", "authentication": "ON_INSTALL" },
      "category": "Developer Tools"
    }
  ]
}

Restart the ChatGPT desktop app or Codex and install Domino Data Lab from the Domino source in the Plugins Directory. This path includes the bundled MCP server (stdio, needs uv).

MCP credentials: the portable mcp.json can't carry secrets, so the server reads DOMINO_API_KEY and DOMINO_HOST from the environment the host passes it. If tools fail with "environment variable not set", see skills/modeling-assistant/SETUP.md → "Register the server manually".

From the public plugin directory

Once published, install Domino Data Lab from the Plugins Directory in ChatGPT or Codex. The directory package is currently skills-only: it doesn't include the MCP server.

---

OpenAI build

scripts/build-openai.sh produces the ZIP you upload at the OpenAI plugin submission portal:

scripts/build-openai.sh                               # dist/dominodatalab-openai-skills-only-<ver>.zip
scripts/build-openai.sh --mcp-url https://host/mcp    # dist/dominodatalab-openai-with-mcp-<ver>.zip

The script:

1. Checks that plugin.json is in sync with .claude-plugin/plugin.json and that the listing fields fit the directory limits (scripts/sync-manifests.py --check).
2. Lints every skill (scripts/lint-skills.py).
3. Stages only portable components (plugin.json, skills/, assets/, LICENSE) and zips them with the plugin root at the archive root.

Choose Skills only in the portal for the default ZIP. The With MCP path requires the Domino MCP server to be deployed at a public HTTPS endpoint using Streamable HTTP with OAuth 2.1; that deployment doesn't exist yet. Until it does, ship skills-only.

The listing icon is assets/logo.svg, used for both logo and composerIcon. It is the pinwheel mark from skills/domino-ui-design/assets/domino-logo.svg on a square #2E2E38 tile, because the portal requires square icons of at least 48×48. The build fails if any icon or screenshot path in plugin.json is missing or outside assets/.

---

Versions

The version is YYYY.XYY.N, for example 2026.603.3, and both manifests carry the same string. It's the release year, the oldest Domino line the content supports written as X×100+Y (603 = Domino 6.3 and later, including Cloud), and a release counter. The format is valid semver, which the OpenAI portal requires. Each release is tagged release-YYYY.XYY.N. See CONTRIBUTING.md "Release branches, tags and backports".

Releases up to 2026.6-3.2 used the older YYYY.X-Y.N form of the same scheme; 2026.603.3 is the first in the current form.

.claude-plugin/plugin.json is the source of truth. Never edit plugin.json's identity fields by hand; change .claude-plugin/plugin.json and run:

scripts/sync-manifests.py

---

What's Included

Bundled MCP Server

A vendored copy of the Domino MCP Server provides tools for running Domino jobs, checking job status/results, and syncing files with DFS-based projects.

• Inside a Domino workspace: Fully automatic — authentication uses ephemeral tokens, project info is auto-detected.
• Outside Domino (laptop): Set DOMINO_API_KEY and DOMINO_HOST as environment variables in your shell.

Requires uv.

Skills (30 total)

| Skill | Description | | --- | --- | | domino-workspaces | Jupyter, VS Code, RStudio workspace management | | domino-jobs | Jobs and scheduled jobs execution | | domino-environments | Compute environments and Dockerfile customization | | domino-datasets | Data management, snapshots, and versioning | | netapp-volumes | Enterprise-grade NetApp ONTAP storage with near-instant snapshots | | domino-projects | Git integration and project collaboration | | domino-apps | Deploy web apps (React, Streamlit, Dash) behind Domino's proxy | | domino-app-init | Scaffold a new Domino-ready app with framework templates | | domino-debug-proxy | Diagnose reverse-proxy and routing issues in apps | | domino-ui-design | Domino UI styling for integrated app design | | domino-ui-bootstrap | Bootstrap a Vite + React project on the Domino design system | | domino-extensions | Build and operate Domino UI Extensions | | domino-experiment-tracking | MLflow experiment tracking and model registry | | domino-experiment-setup | Generate MLflow experiment setup code for a project | | domino-genai-tracing | @add_tracing decorator and DominoRun | | domino-trace-setup | Add GenAI tracing helpers and wiring to an agent project | | domino-model-endpoints | Deploy and call model APIs | | domino-model-serving | Model API and registered-model lifecycle over REST | | domino-model-monitoring | Drift detection and model quality tracking | | domino-governance | Policies, bundles, and evidence for model risk governance | | domino-flows | Flyte-based workflow orchestration | | domino-distributed-computing | Spark, Ray, Dask cluster management | | domino-ai-gateway | LLM proxy for OpenAI, Bedrock, etc. | | domino-launchers | Parameterized web forms for self-service | | `domino-mode