> ## 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

> Understand MCP plan requirements, OAuth scopes, revocation, and data handling.

## Plans

MCP access requires an active **Developer** or **Professional** plan. Active trials, legacy accounts within their 14-day grace period, and grandfathered accounts also retain access. Brokerage data requires **Professional** or an active trial.

## Scopes

Grant only the permissions your AI needs on the OAuth consent screen. Reads require `mcp:read`; brokerage reads also require `brokerage:read`. Each write family has a separate permission:

* `connections:write`: link and revoke connections.
* `syncs:write`: create, change, and run syncs.
* `destinations:write`: manage destinations.
* `categories:write`: manage categories and mappings.
* `rules:write`: manage rules and rulesets.
* `events:write`: manage event destinations and redelivery.

The server filters the tool list to your scopes and refuses calls without the required permission. Enabling the [admin toolset](/docs/mcp/tools#default-and-admin-toolsets) does not bypass these checks. API-key management is not available over OAuth; API-key clients need the appropriate key scopes.

Write tools change Redbark configuration and workflows, not your bank's payment instructions.

## Authentication and revocation

Hosted AI clients use OAuth access tokens and refresh them without storing your Redbark API keys. You can revoke a connected app from [Settings > API & MCP](https://app.redbark.com/settings/api-mcp). To change scopes, disconnect and reconnect the client, then grant the required permissions.

API keys remain supported for manual setups. See [setup](/docs/mcp/setup) for client configuration.

## Data handling

Banking data is returned to your AI client when it calls a tool. Reads may pass through Redbark's short-lived, encrypted connection gateway cache. Financial payloads are not written to the primary database or logs. See [caching](/docs/api-reference/overview#caching) for freshness, retention, purge, and opt-out behaviour.

Your AI provider receives tool results, so its own data policies also apply.

## Context argument

Tools advertise a `context` argument asking the AI for a brief statement of the user's goal. The server accepts calls without it. When supplied, it is used for product analytics, not forwarded as an API argument. Keep banking data and sensitive details out of this field.

Tool arguments, results, and error messages are not captured in MCP analytics.