> ## Documentation Index
> Fetch the complete documentation index at: https://redbark.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Access and security

> How Redbark Advisor MCP connections act as a member, what scopes they hold, and how to disconnect them.

## Acting as a member

A connected app acts as the person who connected it.

* It has that person's role in the firm: admin or member.
* Advisor reads the role on every request. If an admin changes your role, the app follows it on its next call.
* Tools only an admin can use are not listed for a member, and a call to one is refused with `admin_required`.
* Every change the app makes is recorded as you, the same as a change you make in Advisor.

An API key is different. It acts as the admin who made it, with the access chosen when it was made.

## Scopes

Access is set per area. Each area has a read scope and a write scope. Write includes read.

| Area | Scopes | Covers |
| - | - | - |
| Firm | `organisation:read`, `organisation:write` | The firm, its name and its branding. |
| Team | `members:read`, `members:write` | Team members and their roles. |
| Clients | `clients:read`, `clients:write` | Clients, their people and portal invites. |
| Checklists | `checklists:read`, `checklists:write` | Checklist templates, the checklists sent to clients and their answers. |
| Documents | `documents:read`, `documents:write` | Documents clients upload and the details read from them. |
| Money | `money:read`, `money:write` | Bank connections, accounts, balances and transactions. |
| Document types | `document_types:read`, `document_types:write` | Document types, their fields and the client facts they fill. |
| Messages | `messages:read`, `messages:write` | Messages between the firm and clients. |
| Reports | `reports:read`, `reports:write` | Reports exported for clients. |
| Webhooks | `webhooks:read`, `webhooks:write` | Webhook endpoints and their deliveries. |

On the consent page you choose one of two levels:

* **Read only** grants every read scope.
* **Read and change** grants every scope.

An API key can also be **Restricted** to the areas you pick.

The tool list only shows tools your scopes allow. A call without the scope it needs is refused with `insufficient_scope`, naming the scope. [Tools](https://redbark.com/docs/advisor/mcp/tools.md) lists the scope for each tool.

## One connection per person and app

* Each person has at most one connection for each app in each firm.
* Connecting the same app again replaces its access with your new choice. The old connection's tokens stop working.
* To change an app from Read only to Read and change, or back, connect it again.
* If you belong to several firms, connect once for each firm.

## Disconnecting an app

Open **Settings > API and MCP** in Advisor. The **Connected apps** table shows each app, the person who connected it, its access, when it was connected and when it was last used.

* Click **Disconnect** to end a connection. The app stops working straight away. It can be connected again.
* A member sees and disconnects their own apps.
* An admin sees every app in the firm and can disconnect anyone's.

Other ways a connection ends:

* **Removing a member** from the firm ends every app they connected, straight away. API keys they made stop working too. See [team](https://redbark.com/docs/advisor/settings/team.md).
* **Deleting the firm** ends every connection.
* **Unused for 90 days.** Your AI client refreshes its sign-in by itself. If it has not been used for 90 days, connect again.

## API keys

* Only an admin can make or revoke a key, in **Settings > API and MCP**.
* Advisor shows a key once, when it is made.
* Revoking a key stops anything using it straight away. This cannot be undone.
* A firm can have up to 25 active keys.

## Rate limits

* Each connection can make 120 requests a minute. Each API key has its own limit of 120 a minute.
* The limit resets at the start of each minute.
* Answers carry `ratelimit-limit`, `ratelimit-remaining` and `ratelimit-reset` headers.
* Over the limit, the server answers 429 with a `retry-after` header in seconds.

## Client data

* Your AI client sees the data its tools return, such as client names, documents, facts, balances and messages. Your AI provider's own data policies apply to that data.
* Grant **Read only** when you only need answers. Grant **Read and change** when you want the app to act for you.
* Some tools send email to clients: invites, checklists, messages, change requests and documents sent back. Check what your AI client plans to do before it does it.
* Tools marked "Cannot be undone" in [tools](https://redbark.com/docs/advisor/mcp/tools.md) are flagged to your AI client as destructive. Many AI clients ask you to confirm before running one.
* An invite link from `create_client_invite_link` signs the client in. Treat it like a password.
* Download links are short-lived: 5 minutes for a document and 15 minutes for a report.
* Disconnecting a client's bank with `disconnect_connection` deletes that bank's accounts, balances and transactions at once.

For how Advisor protects client data in general, see [security](https://redbark.com/docs/advisor/security.md).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.