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

# Authentication

> Use a separate workspace credential for each person or agent.

The API and MCP endpoint accept `Authorization: Bearer <workspace-token>`. Tokens are provisioned by an administrator in the pilot; OAuth and self-serve token issuance are not available.

Never put a token in a URL or a published document. Keep agent tokens in your client's secret store or environment. Revoke them by removing their hash from Bindo's private runtime configuration.

| Permission | Purpose |
| - | - |
| `customers:read` | Customer history, enquiries, drafts and reply rules |
| `drafts:write` | Create or generate reply drafts |
| `knowledge:read`, `assistant:read` | Read approved knowledge and assistant behaviour |
| `integrations:use` | Let generated drafts call staff-approved connected lookups |
| `integrations:read` | Inspect connected servers and tool permissions |
| `integrations:write` | Configure servers and discover tools; agents cannot approve tool permission grants |
| `billing:read` | Read subscription state |

Configuration also has separate scopes for knowledge, assistant, enquiry, channel and reply-rule changes. Use explicit scopes; a wildcard agent token does not grant management permissions.

> [!WARNING]
>
> Use dedicated `agent` credentials for external agents. Sending replies, granting tool permissions, Checkout and billing-portal access require a human operator. MCP calls cannot exercise these privileges even with an operator token.


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