Documentation
Connecting your tools (MCP / OAuth)
An assistant that can only talk is useful; an assistant that can do things is a product. AI Assistant lets bro call your tools — through your own MCP server — so it can look things up and make changes in your systems, on behalf of your users, under rules you set.
MCP (the Model Context Protocol) is an open standard for exposing tools to an AI. If your systems already speak MCP, bro connects to them the same way it connects to anyone's — no platform-specific glue.
How bro acts on your users' data
When your assistant needs to do something, it calls a tool on your MCP server. Two things make that safe:
- bro tells your server who the user is — as a short-lived, cryptographically signed token your server verifies. It can't be faked or replayed, so your server always knows exactly which of your users a request is for.
- Your server enforces the scope. Because your MCP knows the user, it returns and changes only that user's data. bro never sees more than your server hands it.
Permission tiers
Every tool you expose is assigned a tier, so the assistant can only reach what's appropriate for who's asking:
| Tier | Who it's for | What it allows |
|---|---|---|
| Public | Anyone, including guests | Safe, non-personal reads — product info, general help. |
| Identified | A signed-in user | Reads and actions on their own data only. |
| Delegated | A signed-in user, for actions that need their authorization | The same, for operations you want explicitly user-authorized. |
| Confirm | Any writing action you mark | The user must confirm before it runs — the full action is shown first. |
Destructive or sensitive operations are held behind confirm: bro surfaces exactly what it's about to do and waits for a human "yes." Nothing writes silently.
Setting it up
The Integration section in your AI Assistant Console is the source of truth for this setup. It is generated from the selected tenant's live configuration, so its URLs, identity values, snippets, and release checklist stay current — there is no separate handoff document to keep in sync.
- Open Console → Integration and select the tenant you are configuring.
- Follow its MCP step to Connections, add your server URL and credential binding, then run the probe.
- Review every discovered tool and assign its Public, Identified, Delegated, and Confirm policy deliberately.
- Run preflight and publish the tenant revision. A failed probe or incomplete identity contract blocks publication instead of producing a partly connected assistant.
The Integration section also provides a copy-ready automation brief and the AI Assistant management MCP endpoint for teams that want an agent to perform the same tenant-scoped workflow. See Getting started.
Users connecting their own accounts
Per-user OAuth delegation is available for white-label connectors. Use it when each user must authorize their own account with a service you do not run. Configure the connector's OAuth authorization, token, and revocation metadata in Connections, then publish its delegated tools.
Delegated account tools support two explicit modes:
- Signed customer actor token — recommended when your product has already verified the visitor. AI Assistant mints a short-lived, tenant-and-connector-bound actor token for your MCP server, so account access is automatic and the customer does not see a second sign-in or consent prompt.
- OAuth per signed-in customer — use this when a separate authorization grant is intentional. The assistant shows Authorize account tools once, uses the OAuth 2.1 authorization-code flow with PKCE, and binds the server-side grant to the tenant, connector, and verified end user.
Both modes fail closed. Your MCP server derives the customer only from the verified bearer and never trusts an account identifier supplied in tool arguments. OAuth grants can be revoked from the account menu; signed actor tokens expire in at most five minutes and are pinned to one connector.
Next
- Getting started — where your MCP is connected.
- Governance & model policy — enable tools and set confirm rules.
- The assistant experience — how actions appear to a user.