AI Agent Access — Connect Claude, Cursor & ChatGPT
Let external AI tools your team already uses — Claude Desktop, Cursor, ChatGPT connectors, and any other MCP-compatible app — read and act on tickets in this workspace, using the same role-based permissions the connecting person already has. This runs on the Model Context Protocol (MCP); QueueDesk itself is the MCP server your team connects to.
This is the opposite direction from Connecting Integrations & Custom MCP Servers — that page is about QueAssist calling out to your other systems; this page is about external AI apps connecting in to QueueDesk.
1. Turn it on (Settings → QueAssist)
- •Off by default, per workspace. Go to Settings → QueAssist (
/admin/settings/ai) and find the AI Agent Access card.
- •Flip Allow AI agent connections on.
- •This is a single org-wide switch — turning it off immediately blocks every external AI tool from connecting, and anything already approved stops working the moment it makes its next request.
2. What a connected app can do
Every action a connected app takes is scoped to the connecting person's own QueueDesk role — an AI app never gets more access than the human who approved it already has, and it can never act as a different org than the one they picked when connecting.
| Role | What their connected app can do |
|---|---|
| Employee | Create a ticket, list and check the status of their own tickets, add a comment to their own ticket, search the knowledge base, browse the service catalog. Can never see anyone else's tickets. |
| Agent / Admin / Owner | Everything an employee can, plus: search across every ticket in the org, view full ticket detail (including internal notes), update status/priority/category/assignee, add an internal note, resolve a ticket, generate an AI summary of a ticket's history, and generate a KB-grounded draft reply. |
Tickets created or updated this way trigger the exact same automation as the normal product — assignment rules run on creation, resolve/reopen/reassign emails send on status changes, and CSAT surveys still go out on resolution. From the rest of QueueDesk's point of view, there's no difference between a ticket touched by a person in the UI and one touched by their connected AI app.
3. Connecting an app (any teammate can do this — owner, admin, agent, or employee)
1. In the AI app (Claude Desktop, Cursor, ChatGPT, or any other MCP client), add a new MCP server / connector.
2. Enter the server address: https://queuedesk.com/api/mcp
3. The app will redirect you to sign in to QueueDesk — enter your work email and use the link that's emailed to you, or sign in with your password if your account already has one.
4. You'll land on a consent screen showing the app's name and exactly what it's asking for. If you belong to more than one QueueDesk workspace, a Connect as workspace dropdown appears so you can pick which one this app should act on — most people only belong to one workspace, so most people never see this step.
5. Select Allow access. You're redirected back to the AI app, now connected and ready to use.
4. Managing connected apps (Settings → QueAssist)
Once at least one person has connected an app, a Connected apps list appears under AI Agent Access, showing for each connection:
- •The app's name and icon
- •Who connected it, and the date
- •A Revoke button
Revoking disconnects that app immediately for everyone who approved it under this workspace — its access token stops working right away, on its very next request.
5. Safety and visibility
- •Every call an external AI app makes shows up in the Tool-call audit log at the bottom of Settings → QueAssist — the same table that logs QueAssist's own outbound tool use — tagged External AI app (MCP) in the Source column so you can tell it apart from your team's own agentic activity at a glance.
- •Connections are rate-limited per person, so a misbehaving or misconfigured app can't hammer your workspace.
- •The Emergency stop kill switch does not cover this feature — it only pauses QueueDesk's own outbound agentic actions (agentic resolution, playbooks, integration tool calls). To fully stop external AI apps from reaching your workspace, turn off Allow AI agent connections here instead.
For outbound identity rules (QueAssist calling Google, Microsoft, or your custom server on tickets and in portal chat), see Identity Scoping & Safety.
Next: read QueAssist Settings & Usage Quota for the master AI switch and emergency stop, Identity Scoping & Safety for who QueAssist can act on, or Connecting Integrations & Custom MCP Servers for the outbound direction — QueAssist calling out to Google Workspace, Microsoft Entra, or a server of your own.