Cloverleaf

Yes, and that is its purpose. Cloverleaf's directory listing states Claude looks up colleagues in your organization and pulls their assessment-based insights. The boundary is each colleague's own Cloverleaf visibility setting, not yours. Someone set to Organization is readable by everyone in the org; someone set to Restricted is readable only by people on their approved list.

They consented to the assessment and to a visibility level. Cloverleaf's visibility article describes three levels governing who can see coaching insights about you, written before MCP existed and framed around people. No Cloverleaf document we found asks a colleague whether their profile may be read by a model. Their coworker's OAuth grant is what carries it there.

It returns raw material and the model is instructed to hide it. Cloverleaf's published skill tells the model never to expose scores, framework names or internal ids in user-facing text, and its worked example of what not to print is a trait code, a numeric score and an assertion id. That is a presentation convention in a prompt, not a limit on what the API sends back.

Anthropic's directory lists six. Cloverleaf's own published coaching skill names a seventh, renderFeedback, routed for a review or hard message or recognition. We could not read the live tool list because the endpoint requires authentication, so six is what Anthropic advertises and seven is what the vendor's own instructions assume exists.

One tool writes: startCoachingPlan. Cloverleaf documents that creating a Coaching Focus schedules recurring coaching nudges over a multi-week run and names the specific colleagues the plan is built around. The published skill requires the model to pass confirmedWithUser true and to ask in plain language first. Nothing in the six deletes anything.

No, and the fan-out is not admin-shaped here. Cloverleaf's Organization Reporting dashboard is Org Admin only, but it is a web page with no MCP tool behind it, and it reports adoption counts rather than assessment content. Every MCP user gets the same six tools. What widens reach is a colleague choosing Organization visibility, not a role.

Six scopes, of which two are real. Cloverleaf's resource descriptor advertises openid, profile, email, offline_access, mcp:read and mcp:full. Only the last two govern anything, and mcp:full names no object and no verb, so granting it grants whatever the server implements now and later. Cloverleaf's consent screen phrases it as acting on coaching data on your behalf.

Two separate switches, in two products. A Cloverleaf organization admin must enable the Model Context Protocol card under Organization Settings, Configuration, Integrations, which applies org-wide. On Claude Team and Enterprise plans a Claude Owner must also add Cloverleaf for the organization. Cloverleaf states both are required and that neither substitutes for the other.