# Connecting Claude.ai and ChatGPT

Claude.ai and ChatGPT connect to ResponseHub as a **custom connector**. You sign
in with your ResponseHub account and approve access, and the connection keeps
itself alive from there — there is no token to create, copy or store.

## Add the connector

1. In Claude.ai or ChatGPT, open your connector settings and choose to add a
   custom connector.

2. Paste the ResponseHub server URL:

   ```
   https://app.responsehub.io/mcp
   ```

3. Sign in to ResponseHub when prompted, and approve access.

The assistant discovers everything else it needs on its own — it registers
itself and negotiates the sign-in flow without you filling in a client ID or
secret.

## What you are approving

The approval screen is a ResponseHub page, headed **Connect *(the assistant)* to
ResponseHub**. It shows:

- **Connecting as** — the email address you signed in with, and the account the
  connection will be bound to.
- **This will let it:** — a plain-English list of what the assistant is asking
  for, generated from the access it requested.
- **Authorize** and **Cancel**.

Access is granted for the account shown and expires after two hours, renewing
automatically for as long as the connection stays in use. That means an idle
connector goes quiet on its own, and one you use daily never asks you to sign in
again.

**Only authorise assistants you trust:** Approving a connector gives it your level of access to the account named on the
screen. Read that line before you approve — particularly if you belong to more
than one account.

## What the assistant is allowed to do

Connectors ask for one of three levels of access:

| Access | What it covers |
|---|---|
| **Connect to ResponseHub on your behalf** | Full access — read and write, bounded by your account role. This is what a client gets when it does not ask for anything narrower. |
| **Read your questionnaires, RFPs, knowledge base items and source documents** | Read only. |
| **Create, update and delete questionnaires, RFPs, knowledge base items and source documents** | Write, alongside read. |

Most connectors ask for the first, so assume a new connector can write unless it
says otherwise. If an assistant only ever needs to look things up, and it lets
you choose, pick read-only access.

Whatever it asks for, it can never exceed your own permissions: a read-only
account role stays read-only through a connector. See
[permissions and safety](/help/mcp/permissions-and-safety/).

## Which account you connect to

A connector is bound to **one** account — the first account your user belongs
to. There is no account picker in the approval flow, so if you belong to several
accounts, check the name shown under **Connecting as** before approving.

To work in a different account from a web assistant, use an API token instead:
tokens are created against a chosen account. See
[connecting Claude Code and Cursor](/help/mcp/connecting-claude-code-and-cursor/).

## Removing a connector

Remove or disconnect the connector in Claude.ai or ChatGPT — that is the switch
that stops it being used. ResponseHub does not have its own screen for revoking
a connector's access; if you need a connection cut off at the ResponseHub end,
contact support.

## Next steps

- [What connected assistants can do](/help/mcp/tools/)
- [Reading answers and evidence](/help/mcp/answers-and-evidence/)
- [Permissions and safety](/help/mcp/permissions-and-safety/)