The MCP connector is currently in live beta. We're rolling it out gradually, so it may not be available in your dashboard yet. If you'd like to take part, contact us at [email protected]. As a beta feature, some details described here may change as we refine it.
The ZippyAssist MCP connector lets you connect an AI assistant, such as Claude or ChatGPT, to your ZippyAssist data. Once it's connected, you can ask questions in plain English — like "Which machines had the most refund requests this month?" — and get answers drawn straight from your data, with no exports or spreadsheets.
MCP (Model Context Protocol) is an open standard that lets AI assistants connect to other software. The ZippyAssist connector is read-only: your assistant can look things up, but it can't change anything.
This article covers what your assistant can see, how to connect it, questions to try, and how to manage connections.
What your assistant can look up
Area | What your assistant can tell you |
Dashboard figures | Counts and trends for help requests, refunds, feedback, and voice messages — the same figures as your dashboard home page — by day, week, or month, or broken down by machine, location, account, issue, team member, or hour of day |
Help requests | Requests by status, machine, location, customer, or assignee, with the AI's issue category and summary. It can read any request in full, including its history and notes. |
Refunds | Amounts, statuses, payout methods, and decision notes |
Feedback | Ratings and comments |
Voice messages | Transcripts, translations, and AI summaries |
Escalations | Escalations with their risk levels and call history, if you use Escalations |
Customers | Contact details, history, labels, notes, and ZippyDesk conversations if you use ZippyDesk |
Equipment | Machine details, locations, recent activity, and when each machine last had a submission |
Request links | Links sent to customers, and whether they were opened and completed |
Your setup | Team members and roles, notices, and how your account is set up — statuses, request types, labels, and your refund and escalation rules. Useful for questions like "Why was this refund approved automatically?" |
Read-only by design
Your assistant can't close, reply to, assign, approve, refund, block, create, or edit anything. If you ask it to, it will say so and give you a link to the record in your dashboard, where you can take care of it.
Your permissions still apply
Your assistant sees what you can see. If your user role can't view something in the dashboard, your assistant can't see it either.
Before you start
The connector needs to be turned on for your organization. The ZippyAssist team does this — during the beta, email [email protected] to request access.
Anyone on your team can connect. Once it's on, users with the Owner or User role can each connect their own AI assistant. See Are there different roles for users?
You'll need an AI assistant that supports custom MCP connectors. ZippyAssist provides step-by-step setup for Claude.ai and Claude Desktop, Claude Code, ChatGPT, Cursor, and VS Code. Whether custom connectors are available can depend on your plan with that provider.
Connecting your AI assistant
Click your name at the top of the sidebar, then choose Integrations → AI Agents.
Click Add AI agent and choose the assistant you use. ZippyAssist shows the connector URL along with setup steps for that assistant.
In your AI assistant, add a custom connector and paste in the connector URL.
When prompted, sign in to ZippyAssist and approve the connection. If you have access to more than one organization, choose which ones your assistant can read.
Start a new chat with the connector turned on, and ask your first question.
Tip: Ask your assistant to list the ZippyAssist tools it can use. It's a quick way to see everything it can look up — and the list grows as we add more.
Questions to try
To find out | Try asking |
How busy things are | "How many help requests did we get last week, and how many are still open?" |
Where the problems are | "Which five machines had the most refund requests in the last 30 days?" |
What needs attention | "Show me pending refunds over $10, oldest first." |
What customers are reporting | "What were the most common issues customers reported this month?" |
How customers feel | "Which locations have the lowest feedback scores this quarter?" |
Trends over time | "Chart our help requests by week for the last three months." |
A customer's history | "Summarize everything [customer name] has sent us." |
Quiet machines | "Which machines haven't had a submission in 60 days?" |
Team workload | "Who on my team closed the most help requests this week?" |
Why something happened | "Why was refund [reference] approved automatically?" |
Tips for better answers
Say the time period you mean. Counts and summaries cover the last 30 days unless you ask for something different. Searches and lists aren't limited to a window, so you can ask about older records without them being quietly dropped.
Be specific. Naming a machine, location, account, or customer gets you a sharper answer.
Ask for links. Every record your assistant finds comes with a link to open it in your dashboard.
If you've connected more than one organization
By default your assistant reads every organization on the connection, and answers name which one each figure came from.
You can ask it to focus on one — "only look at Northside from now on" — and that choice sticks, even in new chats, until you ask it to go back to all of them. It isn't a one-off filter on a single question. If later answers look lower than you expect, check whether your assistant is still narrowed to one organization, and ask it to go back to reading all of them.
Good to know
Sticker scans aren't recorded, so the last activity your assistant can see at a machine is its last submission.
Texts and emails sent to customers aren't stored, so your assistant can't see them. It can see ZippyDesk conversations, refund decision notes, and request links.
Issue categories, sentiment, and risk levels come from ZippyAssist's AI analysis of each help request and voice message.
Managing connections
To see which AI assistants are connected, click your name at the top of the sidebar, then choose Integrations → AI Agents.
Each connection shows:
Column | What it shows |
Client (self-reported) | The assistant's name, with its web address underneath |
Connected by | Which team member set it up (shown to Owners, who see the whole organization) |
Connected on | When the connection was approved |
Last used | When that assistant last read anything, or Never |
The client name is self-reported. An assistant tells ZippyAssist its own name and web address when it connects, and nothing verifies either. Treat them as a hint for working out which connection is which — not proof of what the assistant is. If you see one you don't recognise, remove it.
Who sees what
Role | Can view | Can remove |
User | Their own connections | Their own connections |
Owner | Every connection in the organization | Any connection in the organization |
Removing a connection
Open the actions menu on a connection and choose Remove, then confirm. That assistant immediately loses access to this organization, and the connection disappears from the list. If it was connected to other organizations too, it can still read those.
Removing it here doesn't tidy up the other end — the connector may still be listed in your AI assistant's own settings, where you can delete it too.
Your data and your AI provider
When your assistant looks something up, the results are sent to your AI provider — for example, Anthropic for Claude or OpenAI for ChatGPT — so it can answer your question. Those results can include customer names and contact details. Your provider's terms and your account settings with them apply, so make sure the assistant you connect is one your organization is comfortable using with customer data.
Related articles
See also API Keys (for building your own integrations), About webhooks, and Managing Multiple Dashboards.
If you have any questions, use the chat widget at the bottom of the screen or email us at [email protected]!