Cursor team marketplace: rules with named owners

Give every shared rule, skill, and MCP connector a named owner, and your Cursor catalog turns from sprawl into something a teammate can actually trust. A Cursor team marketplace is the shared catalog of rules, skills, and MCP connectors your team standardizes on, with one accountable person per entry. Cursor makes it easy to add capability fast. The owners are what keep that speed safe.

Here is the failure mode the owners prevent. A risky demo lands on main. The merge retro finds a dependency nobody listed: a connector that arrived as a convenience and quietly became policy. By then no one remembers who approved it or why. That is an organizational gap, not a technical one, and you fix it with names and dates, not more tooling.

Write a review receipt into every PR

The fastest way to make reviews answer "why this?" is to put the answer in the PR body. Cursor Blame can tell you which line came from an agent run, but attribution without a story is just trivia.

A review receipt is three things in the PR description: the parent intent, the child scopes, and the verification transcript. With those, a reviewer who was not in the session can trace the whole change without replaying your chat.

Make it a habit, not a hero move. If the receipt is empty, the PR is not ready, the same way an empty test run is not a green build.

Give every connector a steward

MCP connectors multiply faster than ownership, because hosts approve new capability faster than teams assign someone to watch it. That is how a connector becomes anonymous infrastructure.

The fix is a steward: a named human plus a rotation schedule written next to each MCP entry. When a connector's scope changes, there is one person accountable for the change and the rollback. You can read how connectors are defined in the Cursor MCP docs before you assign owners to them.

Stewards also kill the slow drift. A connector with a name on it gets retired when the project ends. One without a name lives forever.

Put your rules on a glob diet

Broad .mdc globs feel tidy and quietly hurt you. When everything matches, agents get contradictory guidance and reviewers cannot tell which file is in charge. Everything-is-active ends up meaning nothing-is-trusted.

A glob diet is simple: split rules by folder concern, and ban the mega-glob unless it ships with explicit exceptions. Then a reviewer can name the governing rule file without a scavenger hunt.

Pair the diet with expiry. Demos ship fast, temporary prompts become defaults, and shortcuts calcify because no one set a retire-by date. Put an owner and a date inside the rule comment so cleanup is scheduled, not heroic.

Here is a small starting rule you can paste and adapt:

---
description: Delegation boundary snapshot (adapt globs to your repo)
globs:
  - "src/**"
alwaysApply: false
---

- Cursor: keep scopes explicit in `.mdc`; forbid undeclared MCP domains.
- Claude Code: cite `CLAUDE.md` precedence before expanding bash scope.
- Codex: keep `AGENTS.md` carrying replay-friendly verification notes for CLI runs.
- Owner: <name>  Retire-by: <date>

Check the receipt before you approve

Approve a marketplace entry when its receipt is complete, not when its demo looks good. These two artifacts make that judgment fast. The table is the question a reviewer asks; the checklist is what the PR author pastes.

Scope receipt:

• Verification command output is pasted or linked.
• Forked agent work lists parent and child responsibilities.
• Red-folder paths got explicit human acknowledgement.
• Scopes in the PR body match the folders in the diff.

None of this replaces architecture judgment. Agents speed up execution; they do not own the decision.

Start with one receipt

Pick your noisiest connector or your broadest glob and give it a named owner today, then make the review receipt non-optional on your next PR. If you want the drills behind these patterns, the cluster lives at Cursor subagents and skills.