Connect Mortar to Claude

Mortar can connect to Claude — and to ChatGPT, Cursor, Windsurf, and any other MCP-compatible AI tool — as a connector, so you can ask questions about your studio's projects and budgets, and (once you approve it) make changes, from outside Mortar itself.

Add the connector in Claude.ai

The fastest way: open Settings → Connected apps inside Mortar and click Add to Claude.ai — it opens Claude with Mortar's connector details already filled in; you just review and confirm.

To add it by hand instead:

In Claude.ai, go to Settings → Connectors → Add custom connector.
For the URL, enter your Mortar MCP address (also shown on Settings → Connected apps inside Mortar) — it looks like https://mcp.mortar.design.
Claude will redirect you to sign in to Mortar and choose which studio to connect. Review the permissions requested (read-only to start) and click Allow.
Claude can now answer questions grounded in your real Mortar data — try "what's my budget status on [project name]?" or "list my active projects."

The same steps work on Claude desktop and the Claude mobile apps — connectors you add in one place are available everywhere you're signed in to the same Claude account.

Add the connector in Claude Code

claude mcp add --transport http mortar https://mcp.mortar.design

This opens the same sign-in and studio-selection flow in your browser; once approved, Claude Code can call Mortar's tools directly from your terminal.

Add the connector in ChatGPT

In ChatGPT, go to Settings → Connectors → Create (or Advanced → Developer mode, depending on your plan).
Give it a name (e.g. "Mortar") and enter your Mortar MCP address (https://mcp.mortar.design).
Complete the same Mortar sign-in and studio-selection flow as above.
ChatGPT can now search your Mortar data (projects, rooms, items, suppliers, invoices, client feedback) and pull up the full detail behind a result — including from deep research, which uses the same search/fetch pair.

Other clients (Cursor, Windsurf, and anything else that speaks MCP)

Mortar's connector is a standard MCP server over Streamable HTTP with OAuth 2.1 (dynamic client registration or a client ID metadata document — whichever your client supports) — any compliant MCP client can connect the same way. Two examples:

Cursor — add to your mcp.json:

{
  "mcpServers": {
    "mortar": { "url": "https://mcp.mortar.design" }
  }
}

Windsurf — add to mcp_config.json (Windsurf calls the field serverUrl, not url):

{
  "mcpServers": {
    "mortar": { "serverUrl": "https://mcp.mortar.design" }
  }
}

Either way, the first tool call opens a browser tab for the same Mortar sign-in and studio-selection flow described above.

What Claude (or ChatGPT, or any connected client) can see and do

Reading is always available: project and room lists, budget summaries, item detail, suppliers, invoices, quote requests, client statements, recent activity, a summary of client comments and reactions on a project, alternative-item suggestions drawn from your own catalogue, and a draft (never sent) client email for an approval nudge, a delivery update, or an invoice reminder — the same data you'd see in the app, plus a general keyword search across all of it. The client-feedback summary shows your private Notes only to a studio owner or admin; anyone else sees just what the client themselves can see. Nothing it reads leaves your studio's own data; a connection can only ever see the one studio you picked when you connected it.

If the connection also has write access (you'll see this scope named on the consent screen when you approve it), it can do two different kinds of things, exactly the same split the in-app AI assistant already follows:

Changes it can make right away — no approval needed. Create and update projects, rooms, items, mood boards, draft invoices, suppliers, catalog services and studio branding: creating a room or item, updating an item's price/status/quantity, setting a room budget, picking a client's preferred option in an alternatives group, adding items to a draft mood board or auto-generating a concept board, adding items to a draft invoice, drafting (never sending) an invoice, adding a product URL by hand, adding or updating a supplier or a catalog service, cancelling a quote request, updating studio branding, and importing an explicitly-selected set of pins from a Pinterest board (see Importing items from a Pinterest board — it always previews first and never imports a whole board blind). Every one of these shows up on your studio's Activity feed as soon as it happens, same as if a teammate had done it. Deleting an item or a catalog service isn't available to a connected app at all — that stays inside Mortar itself (the in-app assistant, or the app's own UI); nothing outside Mortar can do it, approval or not.

Changes that need your approval first. Anything involving money, an email to someone outside your studio, or an outright delete: sending or resending an invoice, voiding an invoice or a payment, recording a payment, emailing a supplier a quote request or a client a review request, creating or revoking a client portal link, inviting a team member or changing whether one is active, changing studio-wide settings, and deleting a whole project, a room, or a project file. It can only request one of these; it can never approve its own request, and if the underlying facts change before you click Confirm (an invoice total edited, say) the request is refused rather than acted on against stale numbers you never actually saw. When it asks, you'll get a notification ("N actions waiting for your approval") and the request appears on Approvals in the main navigation — review the exact facts shown on the card and click Confirm or Deny. A request raised this way stays open for 24 hours before it expires on its own (there's no live chat for it to remind you in, unlike the in-app assistant's own confirmation cards, which use the same Approvals data but expire after 15 minutes since you're already looking at the conversation).

See The AI assistant & natural-language search for more on how the approval step works in chat.

Managing or revoking a connection

Go to Settings → Connected apps to see every app connected to your studio, when it was last used, and to Revoke access — revoking takes effect immediately, on that connection's very next request.

Troubleshooting

"Couldn't reach the MCP server" (or the connector shows as unavailable/offline) Either Mortar's MCP address isn't mapped to a live server yet, or the connector isn't configured for your Mortar environment — this shows up as Settings → Connected apps displaying "Not configured yet" instead of an address. Contact your Mortar admin; there's nothing to fix on your end.

The consent screen shows the wrong studio, or no studio at all You're signed in to Mortar as a member of more than one studio (or of none). Sign out and back in to the right studio account before approving the connection, or ask whoever manages your Mortar studio to invite you first.

A request I approved is stuck / a request expired before I saw it Approval requests from a connected app (unlike the in-app assistant) expire after 24 hours with no reminder inside the client that raised them — check Approvals directly if you're not sure whether something is still pending.

I revoked a connection but it's still working Revoking takes effect on that connection's next request, not instantly mid-flight — a call already in progress when you clicked Revoke can still complete. Nothing after that will succeed.

FAQ

Does this cost extra? No — connecting Claude (or any other client) doesn't use your Mortar seats or change your plan.

Can it see other studios' data? No. A connection is scoped to exactly one studio (the one you picked when you approved it), the same tenant boundary every other part of Mortar enforces.

Is anything it reads sent anywhere else? Mortar doesn't send your data anywhere beyond answering the connected client's request — what that client does with the answer follows its own provider's terms, the same as any other conversation there.