How to connect ChatGPT or Claude to your school data

1. Open the AI access screen

Open Inkwelly and go to the organisation settings, then AI access. This screen lists every existing connection, what each one is allowed to reach, and when it was last used. Treat it exactly like the staff-login screen: it is where access is granted and, just as importantly, where it is withdrawn. If your school has never created a connection, the list starts empty.

2. Select the school the connection belongs to

Select one school. Every connection resolves to a single school inside your organisation, and Inkwelly enforces that boundary on every request — a connection physically cannot read another school's records. Groups running several campuses create one connection per campus, or use a group-level connection for the eight rollup tools that compare schools without exposing individual student records.

3. Set the modules the assistant may read

Set the read scope module by module: fees, attendance, academics, transport, staff, and so on. Scopes are hierarchical, so granting a whole module covers the tools inside it, and granting a narrower group covers only that group. Grant the minimum that answers your actual questions. Tools outside the scope are filtered out before the assistant ever sees the list, so an unscoped module does not exist as far as the assistant is concerned.

4. Choose whether the connection may write anything

Choose deliberately, and default to no. Of the 228 tools, 180 cannot change anything at all. Write permissions live in a completely separate grant namespace, which means a read scope never quietly turns into a write scope — a connection created before a module gained write tools can never write without being explicitly re-scoped. Most schools run read-only for the first month and add writing later, if ever.

5. Enter the connection in ChatGPT or Claude

Enter the connection details in your assistant as an MCP server. In Claude this is the connectors screen; in ChatGPT it is the equivalent connector setting. Nothing downloads and nothing runs on a school computer — the Inkwelly server is hosted, so there is no local process for your IT team to install, patch or monitor. The same connection works from desktop, web and mobile.

6. Confirm the sign-in that authorises access

Confirm the authorisation prompt. The assistant sends you to a normal Inkwelly login, you sign in as yourself, and you approve the connection — the same pattern you already use when authorising any app. Inkwelly issues the credential only after that sign-in succeeds. Nobody types a password into the assistant, and no shared secret is pasted into a chat window.

7. Verify what the assistant can actually see

Verify the scope before trusting it. Ask the assistant to list what it can do for your school, then ask one question you already know the answer to — last month's collection total, or the attendance percentage for one class. If a number looks wrong, the scope or the academic session is usually the cause, not the data. Fix the scope on the AI access screen and ask again.

8. Review the activity log after the first week

Review the log once real questions have been asked. Every call is recorded against the connection that made it, and each write records the exact record it touched with a readable label, so the school can always answer what an assistant read and what it changed. If a connection is reaching further than the work requires, narrow the scope. If it is no longer needed, revoke it — access stops immediately.