Connect AI clients to Uplevel

In Claude, open Settings → Connectors and choose Add custom connector. Paste the Uplevel MCP URL:

https://api.uplevelre.com/mcp

Claude opens Uplevel in your browser. Sign in as you normally would, review what the connector is asking for, and choose Connect. The Uplevel tools appear in Claude straight away.

Add Uplevel as a remote MCP server from the command line:

codex mcp add uplevel --url https://api.uplevelre.com/mcp

Or add it to ~/.codex/config.toml:

[mcp_servers.uplevel]
url = "https://api.uplevelre.com/mcp"

The first time Codex uses the server it opens the same browser sign-in and consent screen.

Any client that supports remote MCP servers with OAuth works the same way: point it at https://api.uplevelre.com/mcp and let it handle discovery. The server implements Streamable HTTP transport, OAuth 2.1 with PKCE, and dynamic client registration, so no Uplevel-specific adapter is needed.

To try it interactively, run the official inspector, pick Streamable HTTP, and enter the URL:

npx @modelcontextprotocol/inspector

If your client cannot register itself automatically, register once and keep the client_id it returns:

curl -X POST https://api.uplevelre.com/mcp/oauth/register \
  -H 'content-type: application/json' \
  -d '{"client_name":"My client","redirect_uris":["http://127.0.0.1:7391/callback"]}'

The same connection works from a design tool such as Claude Design. Everything you create is in your brand by construction: Uplevel serves your brand kit — colors, fonts, logos, voice, license line, plus the email block catalog and merge tags — and the assistant can pull what a design needs straight from your workspace, like a listing's details and photos or the link to its property site.

• "Design a Just Listed email for 3314 Serenity Lane."
• "Make a fall newsletter in my brand and put it in my templates."

A finished design lands in Marketing → Email → Templates marked "New from Claude Design", ready to review, edit, and use. Nothing sends it until you pick it.

One connection can cover every workspace you belong to — you do not reconnect to switch. When you connect, Uplevel asks which workspaces the assistant can use. Choose All workspaces, which includes ones you join later, or check only the ones it needs. Uplevel starts with just the workspace you last had open.

Ask the assistant to list your workspaces first; it calls the list_workspaces tool and gets back the name, slug, your role, and plan of each workspace the connection can use. From then on, every request names the workspace it applies to, and the assistant can move between them freely:

• "List my workspaces."
• "Show the newest contacts in Acme Realty."
• "Now add a follow-up task for that contact in Northside Group instead."

Uplevel never guesses a workspace. If a request does not name one, or names one you are not a member of or did not allow, the tool returns a clear error instead of acting somewhere unintended — which matters most for anything that changes data.

A connection limited to chosen workspaces also cannot do things that belong to your account rather than one workspace, such as creating a workspace, answering invitations, or managing affiliate links.

A connected assistant acts as you and can never exceed your own access. The same role, plan, and record-visibility rules that apply in the dashboard apply to every request it makes. If a contact is hidden from you, it is hidden from the assistant.

When you connect, the consent screen shows what the assistant asked for:

• Read your workspace data — contacts, deals, listings, tasks, campaigns, and settings you can

already see. Every connection gets this.

• Create, update, and delete data — only changes your role already permits. The assistant gets

this only when it asks for it.

An assistant granted read-only access is refused outright when it attempts a change.

Some actions always wait for you, even with write access: sending email, texts, and broadcasts, placing calls, spending money (plans, domains, phone numbers), publishing, deleting a workspace or contacts in bulk, and changing who has access. The assistant gives you a link to an approval page in Uplevel that shows exactly what it wants to do. Nothing happens unless you approve, and each approval covers that one request. You can also find waiting requests under Settings → Integrations → Connected AI clients.

Connecting an outside account, such as Google, Microsoft, or an email provider, happens in the Uplevel dashboard, not through an assistant.

A connection lasts 90 days. After that, the assistant asks you to connect again.

Open Settings → Integrations → Connected AI clients in Uplevel. Every assistant you have authorized is listed there with when it connected, when it was last used, what it can do, and which workspaces it can use. Disconnecting takes effect immediately: the assistant loses access at once and cannot return until you authorize it again from scratch. Anything it already created stays.

Connections are personal — an assistant acts as you, with your own permissions, in the workspaces you allowed. Workspace owners and admins also see a read-only list of the members' connected assistants that can use their workspace, under Settings → Integrations.

The browser page says the request expired. Authorization links are single-use and last ten minutes. Start the connection again from your AI client.

The assistant says it needs write access. The connection was approved for reading only. Disconnect it in Settings → Integrations and reconnect from a client that asks for write access (the write scope).

The assistant says an action needs your approval. Open the link it gives you, or find the request under Settings → Integrations, and approve it within 15 minutes. Then tell the assistant to go ahead.

The assistant cannot find a workspace. Ask it to list your workspaces and use the slug it reports. Workspace names can be ambiguous; slugs and ids never are. If the workspace is not on the list, the connection does not include it: disconnect it in Settings → Integrations and connect again, choosing that workspace.

Requests start failing after heavy use. Tool calls are rate-limited per user per minute to keep automated activity from overwhelming your workspace. Wait a minute and continue.

Your client will not connect at all. Confirm it supports remote MCP servers over HTTP with OAuth — some clients only support local, command-launched servers, which this endpoint is not.

Still stuck? Contact support and mention the client you are using.