Documentation
Last updated: 22 August 2026
How to get help with Workhr, and how the Claude connector works — what it can read, whose permissions it uses, and how to switch it off.
Every screen in Workhr has a Help button in the bottom right. It opens the article for the page you are on, and it can start a conversation with our team without you losing your place.
If you would rather email: support@workhr.app. For anything security-related, see workhr.app/security — please report vulnerabilities there rather than through general support.
Workhr publishes a remote Model Context Protocol server, so you can ask Claude questions about your own workspace — “which credentials expire in the next fortnight”, “what was our margin at East Tamaki last week” — and get answers from your live data rather than a guess.
Turning it on. In Workhr, go to Settings → Integrations → Claude / MCP and switch the connector on. It is off until an administrator turns it on, and any administrator can turn it back off.
Connecting. In Claude, open Settings → Connectors → Add custom connector, and enter the server URL shown on that same Workhr page. Claude sends you back to Workhr to sign in and approve the connection.
It reads. It never writes. Every tool is a query, and the database transaction is switched read-only before the query runs — so nothing can be written even if a query tries.
It uses your access, not the workspace’s. Approving the connection grants Claude what you can see. Queries run as you, through your own row-level security. A consultant and a client-portal supervisor asking the identical question get different answers, and neither can reach another organisation’s data — that boundary is enforced by the database, not by the connector.
Some things are never readable, whatever your role allows: bank accounts, IRD numbers and tax codes; integration credentials, API tokens and OAuth state; calendar connections; and anything in the authentication, storage or vault schemas.
Every call is logged. The audit log records what was asked and how many rows came back — never the rows themselves — including calls that were refused.
Switching it off is immediate. Permission is re-checked on every single request, so turning the connector off stops it on the next call rather than whenever a token happens to expire.
All six are read-only, and none of them accepts an organisation as a parameter.
The connector uses OAuth 2.0. Workhr publishes protected-resource metadata at /api/mcp-meta/prm naming its authorization server, which supports dynamic client registration, PKCE, and the authorization-code and refresh-token grants. You are never asked to paste an API key into Claude, and no long-lived shared secret exists to leak.
How we handle personal information is set out in our privacy policy, and the terms of use in our terms. Our security posture, including how to report a vulnerability, is at workhr.app/security.