Skip to content

Choose Plugin or CLI

Archcore has two entry points. Both read and write the same .archcore/ directory, but they meet you at different levels.

  • Using Claude Code, Cursor, or Codex CLI? Start with the Plugin. You get skills, tracks, built-in agents, and guardrails without extra wiring. (Codex CLI also has a CLI-only fallback if you’d rather not enable the plugin’s plugin_hooks feature flag.)
  • Using any other agent (Copilot, Gemini CLI, OpenCode, Roo Code, Cline), or scripting in CI? Use the CLI directly.
  • Just want to understand the model first? Read the shared concepts.
PluginCLI
Install methodInstall CLI globally, then add the plugin from the marketplaceSingle binary via curl / PowerShell
CLI prerequisiteRequired (same one-line installer as standalone CLI)Required
HostsClaude Code, Cursor, Codex CLI (Copilot planned)All 8 supported agents
Interface7 slash commands (/archcore:init, /archcore:context, /archcore:capture, /archcore:plan, /archcore:decide, /archcore:audit, /archcore:help)MCP tools + natural language
Multi-step workflows/archcore:plan --track and /archcore:decide continuations orchestrate document chains automaticallyYou drive them manually through MCP
GuardrailsMCP-only writes, post-mutation validate, cascade warningsMCP available; guardrails are opt-in via hooks
Best forDay-to-day use inside Claude Code or CursorScripting, CI, agents without plugin support

Both paths share the same .archcore/ directory and the same MCP server backed by the globally-installed archcore binary. Switching paths — or using both at once — requires no migration.

  • CLI is the compiler. It reads .archcore/, builds the graph, speaks MCP.
  • Plugin is the runtime. It applies that graph inside Claude Code, Cursor, and Codex CLI — skills, hooks, workflows.

More depth in Mental Model.