What does archcore init create?

A .archcore/ directory with templates and config for 19 document types in three layers: vision (PRD, idea, plan, RnD, MRD, BRD, URD, BRS, StRS, SyRS, SRS), knowledge (ADR, RFC, rule, guide, doc, spec), and experience (task-type, CPAT).

Which AI agents does the CLI support?

Eight over MCP: Claude Code, Cursor, Gemini CLI, GitHub Copilot, OpenCode, Codex CLI, Roo Code, and Cline (manual setup). Five of those also get session hooks: Claude Code, Cursor, Gemini CLI, Codex CLI, and GitHub Copilot. Run archcore mcp install or archcore hooks install to wire each one up.

Do I need any external services?

No. Standalone binary. Everything in .archcore/ stays in your repo: no servers, databases, accounts, or external dependencies.

Should I install the plugin instead?

If you use Claude Code, Cursor, Codex CLI, or GitHub Copilot CLI, the plugin adds intent-based slash commands on top of the same CLI. Install the CLI on its own when you want the raw context layer or work with another MCP-capable agent.

Git-native project context for every AI coding agent.

The CLI is a single cross-platform binary that creates a .archcore/ directory in your repo, wires up MCP and session hooks, and exposes 19 typed document categories. Vision: PRD, Idea, Plan, RnD, MRD, BRD, URD, BRS, StRS, SyRS, SRS. Knowledge: ADR, RFC, Rule, Guide, Doc, Spec. Experience: Task Type, CPAT.

Each document is markdown with YAML frontmatter, versioned alongside your code. Documents have explicit types and named relations (implements, extends, depends_on, related), so the agent can navigate the dependency graph instead of grepping a flat instruction file. Status fields and timestamps are stored in frontmatter, so reviewers can scan a directory and see what is accepted, draft, deprecated, or superseded.

Works with 8 AI coding agents today over MCP: Claude Code, Cursor, Gemini CLI, GitHub Copilot, OpenCode, Codex CLI, Roo Code, and Cline (manual setup). Five also get session hooks — Claude Code, Cursor, Gemini CLI, Codex CLI, and GitHub Copilot; OpenCode is never wired for hooks because its hooks are JavaScript plugins that cannot be written declaratively. Two commands wire each one up — archcore mcp install registers the local MCP server, archcore hooks install adds session hooks where the host supports them. Claude Code, Cursor, Codex CLI, and GitHub Copilot CLI are also plugin hosts. Anything else that speaks MCP works the same way: the CLI is a local MCP server, not an integration per vendor. The MCP server runs locally as a child process and exposes tools to list, get, create, and update documents during a real session. Hooks pre-load relevant context based on the files in scope, so the agent starts each turn with the right rules and specs already in view.

Already have instruction files? You don't start over. archcore init imports the ones you already wrote (CLAUDE.md, AGENTS.md, .cursorrules, .cursor/rules/*) as typed documents: conventions become rules, the reasoning behind them becomes ADRs, and prose already sitting in docs/ becomes guides, specs, and plans. Path-scoped instruction files keep their scope but gain status, timestamps, and relations, so an agent pulls the guide for the directory it is editing instead of grepping the whole folder.

Install with curl -fsSL https://archcore.ai/install.sh | bash on macOS or Linux, or irm https://archcore.ai/install.ps1 | iex on Windows. Cross-platform binary on amd64 and arm64. Run archcore doctor to verify setup, archcore update to self-update, archcore hooks install and archcore mcp install to wire up the integrations. No Node, no Python, no external services required.

Typical first steps after install: run archcore init to scaffold the directory and wire your agents, archcore status to see the document tree and health, and archcore doctor to verify MCP registration and hook installation in one shot. Day-to-day work happens inside the agent — capturing decisions, codifying standards, planning features, and detecting stale documentation as the codebase evolves.