Skip to main content
External agents (ChatGPT, Codex, Claude) should call ar_get_started on Platform MCP first, then load skill markdown from the public tree below (no access to the private platform-mcp GitHub repo required). Base URL:
Browse the full tree on GitHub: platform-mcp-skills.

JSON schemas (embedded contracts)

Or call Platform MCP tenantdata_logical_schema and surfaces_schema_catalog — they return the same public URLs.

Typical order

  1. ar_get_started — tenant, playbooks, embedded_schema_tools
  2. Load platform-design SKILL, then tenant-data → workflow → surfaces as needed
  3. workflows_validate with tenantdata_schema before run
  4. For Surfaces apps: publish workflows first, then set workflow_ref to published UUIDs (avoid Import pending in Console)