06-reference/research

claude marketplace free skill paid cta policy

2026-08-24·research-brief·source: deep-research·by Ray Data Co (deep-research synthesis)
macplugin-marketplaceanthropic-policylisting-constraintsfree-skill

One clause governs this, it lives in a policy that only binds Anthropic's own Directories, and it is a primary-purpose test rather than a link ban

The question

"Does the Anthropic/Claude plugin marketplace permit a free skill whose output links to a paid off-platform product, and are there listing/policy constraints on commercial CTAs inside a free skill?" Named open follow-up #2 in [[2026-08-18-free-skill-funnel-paid-conversion]], which recommended shipping MAC's free /dq audit as an instrumented marketplace skill and left the legality of its CTA unresolved.

What we already know (from the vault)

What the web says

Convergences and contradictions

Synthesis for RDCO

The direct answer, stated at the confidence the sources support: nothing in Anthropic's published policy prohibits a free skill from linking to the publisher's own paid off-platform product, and no source found imposes any listing constraint on commercial CTAs specifically. But that is not a permission, it is a silence plus one adjacent prohibition. The only clause that touches commerce is section 4.C, and it does not name links, CTAs, upsells, or pricing. Anyone claiming Anthropic "allows" or "forbids" the MAC CTA is reading a rule that is not written. What is written is a primary-purpose test, and MAC's exposure sits entirely inside it: a free /dq audit is fine as long as it is software that does a real job and happens to end with a next step; it becomes a policy problem the moment it is a thin wrapper whose actual function is to advertise a $350 product. That maps with uncomfortable precision onto the design constraint the parent brief already derived from Freemius on entirely commercial grounds - gate after value is realized, do not ship a trimmed demo that produces nothing worth acting on. The compliance argument and the conversion argument want the same artifact. That is a rare alignment and RDCO should treat it as the design brief rather than as two separate reviews.

The governance boundary, not the clause, is the decision-relevant finding. RDCO holds two publishing surfaces with opposite risk profiles. Publishing /dq to ray-plugins invokes no Anthropic listing review at all - the docs are purely mechanical, the only enforcement is anti-impersonation naming, and ray-plugins is already compliant on that axis. The cost is that the parent brief's entire strategic point was reach into the commons where buyers already are, and a self-hosted catalog delivers approximately none of it against an in-app Discover screen shipping 200+ plugins by default. Publishing to the community or official directory buys that reach and simultaneously imports 4.C plus an unpublished "quality and security standards" bar behind a submission form. This is the same trade [[2026-07-21-anthropic-plugin-marketplace-brigade-strategy]] identified for the Verified badge, and it should be decided once for both: RDCO either rides Anthropic's rails and builds to Anthropic's standards, or it keeps its own surface and accepts that the surface is not distribution.

Observed practice is the strongest evidence available, and it is genuinely strong, but it is not a precedent for MAC's exact shape. A Stripe or Linear connector is a free plugin whose entire function is to reach a paid product the user is already paying for. It is instrumentally tied to that product; it is not a customer-acquisition device for it. MAC's /dq audit is the other thing - a free artifact aimed at people who do not yet own the paid product, whose measured success is click-through. No listing in the surveyed 255 was identified as serving that function, so RDCO would not be violating an observed norm so much as testing an unobserved one. The practical read is that the risk is low and the failure mode is cheap: a rejected submission, not an account action, and the self-hosted fallback remains open either way.

What RDCO should not do is treat this brief as clearance. The Software Directory Policy was reached through a single targeted fetch that returned an extraction rather than the full text end to end, and the submission form's own criteria were never retrieved. Both gaps are closable in under an hour by a human reading two pages. The correct sequencing is: design /dq to stand alone as software with the CTA as a terminal next step rather than as its purpose; publish first to ray-plugins where no review applies and the instrumentation from the parent brief can start collecting the funnel data that does not exist anywhere in this category; and read the full policy plus the submission form before, not after, deciding to submit for directory reach.

Why this is in the vault

This closes open follow-up #2 of [[2026-08-18-free-skill-funnel-paid-conversion]] and removes the stated blocker on MAC's unshipped free /dq audit skill: the CTA is not prohibited by any published Anthropic text, so the go/no-go on shipping the lead magnet as an installable skill now turns on a design choice (does the audit stand alone as software) and a distribution choice (self-hosted ray-plugins with zero review versus directory submission with an unpublished bar), not on a legal unknown. It also supplies the missing commercial-content dimension to the ray-plugins submission decision left open in [[2026-07-21-anthropic-plugin-marketplace-brigade-strategy]].

Open follow-ups

Related

Sources