MCP integration

When the MCP integration feature is enabled for your organization, a toggle appears under Settings → API access that lets an admin allow an AI assistant (for example Claude) to read the organization’s employees, surveys and analytics via MCP (Model Context Protocol).

Turning the toggle on automatically creates a dedicated API key with default, least-privilege scopes: read employees, read surveys and read analytics — never salaries or personal identifiers. The key is shown only once, right after it’s created, together with the connection URL and a ready-to-paste config for an MCP client. An admin who wants MCP to also see salaries can add the employees:read:pii scope manually in the regular API keys panel.

Turning the toggle off revokes the key immediately and requires confirmation — every connected MCP client stops working right away. The key managed by this toggle does not appear in the regular API keys list; it’s managed here only.

MCP instances (advanced). Below the toggle is an advanced section where an admin can create several named MCP instances, each exposing its own subset of the tools (employees, surveys, the three analytics tools, and the onboarding guide below) and each independently switchable to include salaries. That lets, for example, an HR instance read salaries while a general instance sees only employees without salaries. Each instance is assigned to individual users — every assigned user gets their own key, shown once and revocable individually without affecting the others. A personal key also stops working automatically as soon as its user no longer has active access to the organization — when the account is deactivated, removed from the organization, or their employment end date has passed — even if nobody has revoked the key yet. If the person regains access the same key works again; to close it for good, remove the user from the instance. The connected MCP server only shows the tools an instance’s key may actually use; disabled tools are always blocked by the API.

Onboarding guide (own checklist only). The “onboarding guide” tool works differently from the others: the key an assigned user gets is tied to that specific person, and the tool ALWAYS shows only that person’s own onboarding checklist (for example, to a new hire’s AI assistant) — never another employee’s, no matter what is asked for. A regular, non-personal API key cannot use this tool at all.