Experience Design System SDK

Contentful Experiences lets you compose pages and layouts from your own design system components. The @contentful/experience-design-system-cli imports your design system into Contentful. It extracts Component Type definitions from your local codebase, invokes an AI agent to generate Component Definition Format (CDF) definitions, and pushes them to your Contentful space.

Your codebase remains the single source of truth. The CLI analyzes your source files (.tsx, .ts, .jsx, .js, .vue, .astro) using static analysis, then delegates property classification and CDF generation to a coding agent.

Prerequisites

• Node.js 24
• pnpm 10.27.0+
• A coding agent in $PATH — Claude Code, Codex, OpenCode, or Cursor
• A Contentful CMA token — set CONTENTFUL_MANAGEMENT_TOKEN

Quick Start

Install

git clone https://github.com/contentful/experience-design-system-sdk-public.git
cd experience-design-system-sdk-public
pnpm install
pnpm build
pnpm -F @contentful/experience-design-system-cli build

The final step builds the CLI package and symlinks the experiences, exo, and experience-design-system-cli commands alongside your node binary so they're available on your $PATH.

The package publishes three binaries — experiences, exo, and experience-design-system-cli. The docs below use experiences since that is the wizard-oriented entry point.

Setup

experiences setup

The interactive setup wizard installs Node 24 (if needed), verifies pnpm, checks for a coding agent, and persists credentials + agent preferences to ~/.config/experiences/credentials.json. Later commands read those values automatically.

Run

The primary entry point is the import wizard:

experiences import

In an interactive terminal this launches a full-screen TUI that walks you through extraction, AI selection, manual scope review, generation, final review, and push. Credentials and project path are pre-filled from experiences setup. Component generation runs in parallel with credentials entry so the wizard does not block on the agent.

For scripted or CI use, pass --auto-accept-scope plus credentials:

experiences import \
  --project /path/to/your/component-library \
  --space-id $CONTENTFUL_SPACE_ID \
  --environment-id master \
  --cma-token $CONTENTFUL_MANAGEMENT_TOKEN \
  --auto-accept-scope \
  --yes

How it works

The CLI runs your component library through four stages:

1. Analyze — Reads your source files and extracts every component: its name, props, types, and source location.

2. Select — An AI agent reviews the extracted components and decides which ones make sense to expose in Contentful Experiences (buttons, cards, layouts) and which to skip (hooks, context providers, utilities). The wizard then opens a single scope-gate for you to confirm or override that list. The AI auto-filter can be turned off with --no-auto-filter.

3. Generate — An AI agent takes the selected components and produces structured definitions that tell Contentful what each prop is for — whether it holds content, a design token, or interactive state. The wizard then opens a final-review field editor where you can edit names, descriptions, defaults, allowed values, and slot constraints inline.

4. Apply — Shows a diff of what will change in your Contentful space, then pushes (and by default also saves components.json / tokens.json to disk for source control).

The wizard saves a run record after each session. Use experiences runs to list prior sessions (or experiences runs <id-or-path> for a single-run detail view), experiences import --modify <id-or-path> to re-open the wizard pre-populated, or experiences import --push-from-run <id-or-path> to re-push without re-opening the wizard.

When prior runs exist and the wizard is launched without --push-from-run, --modify, or --project, the TUI opens with an interactive run picker so the operator can pick "push", "modify", or "start a new run" up front. Pass --project to skip the picker and go straight into a fresh extraction.

For headless operation alongside an existing checked-in components.json, pass --on-conflict <overwrite|skip|fail> to bypass the interactive save-conflict gate. Use --print-prompt to inspect the generate prompt without invoking the agent (replaces the deprecated --dry-run).

Packages

Package	Description
@contentful/experience-design-system-cli	The CLI + interactive wizard — analyze, review, generate, validate, and push component definitions
@contentful/experience-design-system-types	Shared types and schemas for the CDF and DTCG data formats

Command Reference

Full documentation for every flag and every subcommand lives in packages/experience-design-system-cli/README.md.

Command	What it does
experiences setup	Interactive setup — installs prerequisites and saves credentials + agent
experiences doctor	Health check — verify Node version, credentials, and agent binaries
experiences import	Run the full wizard or a headless pipeline (extract → select → generate → push)
experiences runs	List prior wizard runs, or pass <id-or-path> for a single-run detail view (supports --json, --pushed, --not-pushed)
experiences analyze extract	Scan source files and extract raw component definitions
experiences analyze select	Interactively pick which components to include (standalone JsonEditor TUI)
experiences analyze select-agent	AI agent picks which components belong in Experiences; pass --show-rationale [--json] for read-only rationale output
experiences generate components	AI agent generates CDF definitions from raw analysis
experiences generate tokens	AI agent generates DTCG design tokens from raw token data
experiences apply preview	Read-only diff — what would change in Contentful
experiences apply select	Checkbox TUI to pick a subset of entities to push
experiences apply push	Write component types and design tokens to Contentful; emits webapp view URL
experiences print components	Export generated components to components.json
experiences print tokens	Export generated tokens to tokens.json
experiences print validate	Validate CDF or DTCG files against their schemas
experiences session list	List pipeline sessions (lower-level than runs)