DocsDevelopersClaude Code

Claude Code

Build and deploy RootCX apps and AI agents directly from your terminal. The RootCX skill for Claude Code teaches it how to write manifests, frontends, backends, and agents that deploy straight to your project.

Claude Code with the RootCX skill produces code that is 100% compatible with RootCX Studio. The same 6 knowledge modules power both tools.


Getting Started

1. Install the CLI and sign in

curl -fsSL https://rootcx.com/install.sh | sh
rootcx auth login https://<your-project>.rootcx.com

If you don't have a project yet, run rootcx init to create one from scratch.

2. Add the RootCX skill to Claude Code

npx skills add rootcx/skills

3. Start building

Describe what you want in plain language:

"Build a CRM to manage contacts, companies, and deals with a sales pipeline.
 Add a support agent that can look up contact history and create follow-up tasks."

Claude Code generates the manifest, frontend, backend, and agent configuration using the 6 RootCX skills. Deploy with:

rootcx deploy

That's it. Your app is live at https://<your-ref>.rootcx.com/apps/<appId>/.


The 6 RootCX Skills

Skills are knowledge modules that guide Claude Code when writing RootCX applications. They load automatically and activate contextually.

rootcx-manifest -- Data Contract & RBAC

Teaches Claude Code to write manifest.json:

  • Entity definitions with field types (text, number, boolean, date, timestamp, json, file, entity_link, [text], [number])
  • System columns id, created_at, updated_at (auto-generated, never declared)
  • Relationships via entity_link with references (local entities or core:users)
  • RBAC permission keys (entity.action convention, auto-prefixed with app:<appId>:)
  • Schema sync behavior -- Core auto-diffs and migrates on every deploy

rootcx-sdk-hooks -- React Data Hooks

Teaches Claude Code to use @rootcx/sdk:

  • useAppCollection<T>(appId, entity, query?) -- list, filter, create, update, delete
  • useAppRecord<T>(appId, entity, id) -- single record CRUD
  • useIntegration(integrationId) -- connect, call actions, manage OAuth
  • useCoreCollection<T>(entity) -- read-only access to platform data (users)
  • useRuntimeClient() -- imperative API for event handlers
  • Full where-clause DSL: $eq, $ne, $gt, $lt, $like, $ilike, $in, $contains, $isNull, $and, $or

rootcx-ui -- UI Components & Styling

Teaches Claude Code to build frontends with @rootcx/ui:

  • AppShell + Sidebar layout pattern
  • AuthGate entry point (login/register/OIDC)
  • DataTable with server-side pagination and sorting
  • Forms, dialogs, toasts, and feedback components
  • Dark mode via ThemeProvider + semantic CSS tokens
  • Tailwind CSS v4 styling conventions (bg-background, text-foreground, border-border)

rootcx-backend-worker -- Bun Worker & IPC

Teaches Claude Code to write backend logic:

  • JSON-lines IPC protocol between Core and Bun worker
  • RPC handlers with caller authentication (userId, email, authToken)
  • Background job processing via onJob hook
  • Direct PostgreSQL access via database_url
  • Lifecycle hooks: onStart, onJob, onError, onShutdown
  • 30-second RPC timeout -- use jobs for long-running work

rootcx-rest-api -- Core HTTP API

Teaches Claude Code to call the Core API directly:

  • Collections CRUD: GET, POST, PATCH, DELETE on /api/v1/apps/{appId}/collections/{entity}
  • Bulk operations: POST /bulk and POST /query
  • Integration actions: POST /api/v1/integrations/{id}/actions/{actionId}
  • Job queue: POST /api/v1/apps/{appId}/jobs
  • Bearer token authentication for all requests

rootcx-agent -- LangGraph AI Agents

Teaches Claude Code to build AI agents:

  • backend/agent.json configuration (name, system prompt, memory, limits, supervision)
  • LangGraph createReactAgent with provider-specific LLM SDK
  • IPC bridge for tool calls (query_data, mutate_data, invoke_agent)
  • Supervision modes (autonomous, supervised, strict) with policies
  • SSE invocation endpoint and session management
  • System prompt best practices

How It Works

  1. Skills (6 knowledge modules) -- loaded into Claude Code's context when relevant. They are the single source of truth, shared with Studio's AI Forge.

  2. CLI (rootcx binary) -- the execution layer. Signs in to your Core, scaffolds projects, deploys code, starts workers.

Skills Path Resolution

The CLI finds skills in this order:

  1. ROOTCX_SKILLS_DIR environment variable (if set)
  2. Relative to the rootcx binary (../.agents/skills/)
  3. ~/.rootcx/skills/ (fallback)

What You Can Build

From Claude Code, you can build anything the platform supports:

  • Full-stack applications -- CRM, project tracker, billing dashboard. Manifest + React frontend + backend RPC handlers.
  • AI agents -- customer support bots, data analysts, workflow automators. LangGraph backend with tools and supervision.
  • Integrations -- Slack, Stripe, email connectors. Actions, webhooks, and credential management.
  • Cross-app systems -- agents that query data from multiple apps and delegate to other agents.

All deployed artifacts are identical to what Studio produces. You can start in Claude Code and switch to Studio (or vice versa) at any point.