BrandKit Nexus MCP
Connect Claude or ChatGPT to Nexus with OAuth. Growth and Custom, once released to your workspace.
Claude setup
In Claude, open Settings → Connectors → Add custom connector, then use these values. Claude shows several options; the two that matter are set out below.
Name
BrandKit
Remote MCP server URL
Shown once the connector is enabled for your workspace.
Authentication → Always required
BrandKit requires a sign-in for every tool, so each person who uses the connector signs in to their own BrandKit account and gets their own permissions. Claude detects this automatically.
OAuth client → Use Anthropic's hosted client metadata
Claude's recommended option, and the one to pick. BrandKit reads Claude's client details from the URL Anthropic publishes, so there is nothing to register and nothing to paste. The “register one automatically” option also works if you prefer it.
Additional request headers → leave empty
BrandKit authenticates with the OAuth token alone. Do not add an API key here.
ChatGPT setup
Name
BrandKit
Server URL
Shown once the connector is enabled for your workspace.
SSE fallback URL
Shown once the connector is enabled for your workspace.
Authentication
OAuth
When BrandKit adds a new action
Claude and ChatGPT read the list of available actions once and then cache it. New BrandKit actions are live the moment we ship them, but your assistant keeps showing the list it already has until it reads it again. Nothing needs reinstalling.
- 1Open the BrandKit connector in your assistant’s settings and refresh or reconnect it. This updates the stored list.
- 2Start a new chat. An existing conversation keeps the action list it was given when it started, so a refresh alone will not change what that chat can see.
- 3Ask the assistant to list its BrandKit actions and compare the count with the number below.
The server version above changes whenever the action list changes, so it is the quickest way to tell a stale assistant from a stale server. If your assistant reports fewer actions than this after a new chat, tell us the version it reports.
What the connector can and cannot do
Both assistants use OAuth with PKCE. The connection is bound to the person who signed in — their workspace, their Nexus role permissions and their plan — so an assistant can never see more than that person can see in BrandKit.
- Reading and drafting is available: search records, brief a client, judge stale quotes, summarise the pipeline, draft a follow-up.
- Anything that commits customer-facing work stays inside BrandKit. Preparing and confirming a follow-up, creating invoice drafts, and sales handover are deliberately not exposed, so an external agent cannot both propose and approve the same action.
- Access can be withdrawn at any time. Tokens are re-checked against your workspace release and plan on every call, not only when the connector was added.
Growth customers get early access to new beta AI capabilities first. MCP is closed beta while we refine the safest, most useful Nexus Operator workflows.