What is the MCP-in / CLI-out loop?

MCP-in: you improve an instruction in your editor through the Rubrkit MCP server. CLI-out: once a version is approved, npx rubrkit pull places it into any project. One source of truth, in wherever the agent actually reads it.

Does it work across Claude, Codex, and Cursor?

Yes. The registry is runtime-agnostic; the adapters handle placement so one approved instruction stays in sync across the tools your team uses.

How is a version approved?

A version is approved when it clears the rubric bar you set. The approval and its score travel with the version, so anyone pulling it can see why it shipped — not just that it did.

Can I roll back to an earlier version?

Yes. History is first-class, so you can pull any earlier approved version into any project at any time. Rollback is just another pull.

Will pulling overwrite my local edits?

Pull is explicit and writes to the runtime’s known path. Treat that path as managed by the registry — the approved version is the source of truth — and keep local-only notes outside it.

Registry & sync

Version control for your skills, agents, and prompts.

Improve an instruction in your editor over MCP. Sync the approved version into any project with one command. One source of truth across Claude, Codex, and Cursor.

Connect the MCP server CLI docs

One command

npx rubrkit pull skill/code-review

→ places the approved version in CLAUDE.md, AGENTS.md, or .rubrkit/

The loop

Improve once. Sync everywhere.

MCP-in to improve in the editor, CLI-out to place the approved version wherever the agent reads it — with version history and provenance on every step.

01

Editor

You refine an instruction in your editor.

02

MCP in

Rubrkit grades and tracks the version over MCP.

03

Registry

The approved version becomes the source of truth.

04

CLI out

npx rubrkit pull places it into any project.

Claude

→

CLAUDE.md

Codex

→

AGENTS.md

Tooling

→

.rubrkit/

What syncs, and where

One approved version, placed per runtime.

The registry is the source of truth; the adapters handle the last mile so each tool reads the instruction from the path it already expects. You improve once and place everywhere, with the score and provenance riding along.

Version history with provenance

Every version carries its rubric score, the audit that approved it, and what changed — history with a quality bar, not just a diff.

MCP in

Improve an instruction in your editor over the Rubrkit MCP server, grading as you go.

CLI out

npx rubrkit pull places the approved version into any project, in CI or on a laptop, with one command.

Per-runtime adapters

CLAUDE.md for Claude, AGENTS.md for Codex, and a .rubrkit/ directory for tooling that reads from a known path.

Developers

Bring Rubrkit into your toolchain.

Sync graded artifacts into the repos where your agents run, or connect any MCP client to the same audits, bundles, and rubr_flow tools the web app uses.

CLI

rubrkit on npm

Pull approved artifact bundles into local projects and place them where Codex, Claude, or generic agents expect their instructions.

# Pull approved artifacts into your project
npx rubrkit pull

# Place them where your agent expects them
npx rubrkit pull all --agent claude --yes

View on npm API docs

MCP server

Public

Point any MCP client at Rubrkit and call the same artifact bundle, audit, and rubr_flow tools, authenticated with your Rubrkit API key.

{
  "mcpServers": {
    "rubrkit": {
      "url": "https://rubrkit.com/api/v1/mcp",
      "headers": {
        "Authorization": "Bearer <your-rubrkit-api-key>"
      }
    }
  }
}

Create an API key API docs

FAQ

Answers before you start.

Git versions files; it has no idea what a good instruction is. Rubrkit versions instructions as graded artifacts — every version carries its rubric score, provenance, and the audit that approved it — and syncs the approved version into any project. You get history plus a quality bar, not just a diff.

Version control for your skills, agents, and prompts.

Improve once. Sync everywhere.

One approved version, placed per runtime.

Version history with provenance

MCP in

CLI out

Per-runtime adapters

Bring Rubrkit into your toolchain.

Answers before you start.

Know which instructions are ready to run.

Follow the review loop as it ships.