Claude's Connector Directory Is Open to Anyone Now. Your Business Needs Two Policies.

What changed: the directory is no longer curated-by-default

Anthropic launched the connectors directory in mid-2025 with a handful of named launch partners. On September 25, 2026, that changed: Anthropic opened a developer portal where anyone on a paid Claude plan can submit an MCP connector or a plugin bundle for listing — “there's no partner program to apply to first,” in Anthropic's own words. A submission is either a connector pointed at a remote MCP server, or a plugin bundle hosted on GitHub. Anthropic checks each one before it goes live, and developers get installs, per-tool usage, and error-rate figures once it's published.

That is a genuinely good thing for the ecosystem — it is also the reason “it's in the Claude connectors directory” stopped being a meaningful trust signal on its own the moment the gate opened to everyone. What matters now is which label a given listing carries.

Three labels, and what each one actually promises

Anthropic's own connector verification documentation is unusually direct about what its review tiers do and do not cover:

The escalation from Community to Verified is automatic and developer-initiated by nothing in particular — Anthropic's docs say it “may escalate listings flagged as highly useful,” with no separate application process. A business cannot assume a connector will stay in whichever tier it started in, and cannot assume Verified means audited. It means Anthropic's reviewers ran the tools once and they worked as described.

The other axis: who gets a connector, solved separately

Verification is about the connector. A completely different question — who on your team can reach it, and under what credential — is what Anthropic's Enterprise-Managed Authorization reached general availability for on August 24, 2026. Before it, every employee who wanted a connector ran their own OAuth consent flow, or someone pasted a long-lived API key into a shared config and hoped it got rotated. Under Enterprise-Managed Auth, an admin authorizes a connector once against the organization's identity provider — Okta at launch — and employees inherit access automatically through the IdP groups and roles they already hold. No consent screen, no manual OAuth step, no separate credential for someone to lose track of. Deprovision someone in the IdP and their connector access goes with it, on the same timeline as every other system tied to that identity.

Launch coverage included Asana, Atlassian, Canva, Figma, Granola, Linear, and Supabase, with Datadog, Notion, and Slack added at GA. It is built as an extension to the open Model Context Protocol that any connector — including one your own engineers write — can adopt, not a proprietary Anthropic-only mechanism.

Notice what this does and does not fix. It removes the static-credential problem — the exact gap that made ad hoc API keys the weak point in connector security last year. It does nothing about whether the connector on the other end of that clean, IdP-verified connection is one your business should have authorized in the first place. A Community connector reached through Enterprise-Managed Auth is still a Community connector; the plumbing got safer, the destination didn't get more trustworthy.

The checklist: two policies, not one

• Write down who can approve a new connector for the business — not just who can click “connect” in their own Claude session. An open directory means anyone on the team can find and request a connector you have never heard of.
• Treat the label as a starting filter, not a decision. Verified narrows the field; it does not replace a look at what data the connector touches and who operates the server behind it.
• Default to your identity provider for anything that will see real company data. If Enterprise-Managed Auth supports your connectors and your IdP, use it — it is strictly better than a personal OAuth grant or a shared key, and it makes offboarding automatic instead of a checklist item someone forgets.
• Keep a live allowlist, not a one-time approval. A connector's developer can change its tools after your review, on either label tier. The record of what is connected needs to be something you check, not something you filed once.
• Hold Custom connectors to the highest bar, not the lowest. No one at Anthropic has looked at them. If someone adds one by URL, your business is the only review it gets.

Where Skoor fits

This is the governance layer Skoor builds into every Claude rollout: role profiles with scoped permissions, a written connector allowlist tied to what each profile actually needs, and continuous drift scoring against that profile so a connector that gets added — or quietly changes — doesn't sit unnoticed. The approach, the scoring rubric, and a sample report are public at skoor.ai/solutions and skoor.ai/work. Skoor installs and manages Anthropic's own Claude apps and connectors on business devices — it is not a Claude interface of its own.