AI Connection
Let approved AI assistants use live Sparki context.
AI add-ons
Give an approved assistant the context it needs before it helps.
The AI connection uses Sparki's remote MCP server and OAuth, so answers can use live portal data while access still follows the authorized user and permissions.
The AI connection is for teams that want to use assistants such as ChatGPT or Claude with live Sparki context. Instead of copying details from the dashboard by hand, the assistant can request approved information from Sparki and use it to answer questions, prepare follow-ups, or explain what happened recently.
What You Can Ask
ProjectsGet a quick read on active projects and jump to the right dashboard page when action is needed.
AppointmentsCheck availability, requests, and setup links without digging through each agenda by hand.
Inbox follow-upFind messages, callbacks, requests, and time proposals that may need follow-up first.
You can also ask about customer timelines, recent interactions, documents, bids, workflow answers, configured workflow procedures, and portal statistics when those tools are enabled for the connected AI.
What Sparki Can Share
The MCP connector exposes read tools first. The exact answer depends on the OAuth scopes the connected assistant received and the portal the user authorized.
| Area | What the assistant can look up |
|---|---|
| Projects | Project lists, project titles and descriptions, status, dashboard links, portal links, schedules, workflow links, and pre-validation workflow links. |
| Appointments | Recent appointments, project schedules, upcoming slots, booked appointment counts, appointment customers, open appointment requests, and time proposals. |
| Inbox | Recent interactions, unread/newest filters, outbound filters, request filters, urgent follow-up filters, messages, appointment requests, documents, offers, orders, and workflow responses. |
| Customers | A customer profile by customer ID or email, plus their recent timeline with messages, requests, appointments, documents, offers, orders, and workflow answers. |
| Workflows | Active workflows, linked projects, steps, questions, options, conditions, outcomes, and dashboard links. |
| Statistics | Basic portal counts for inbox interactions, appointments, projects, and customers. |
Customer export is separate from normal lookup. It requires the customers:export scope and explicit approval before customer records are returned.
Good first prompts:
Try prompts like:
"What projects do we have?"
"When can someone book this project?"
"Which inbox items need urgent follow-up?"
"What happened with this customer recently?"
"Explain the procedure in this workflow."
Set Up the Connection
Open the active portal in the dashboard, then choose Add-ons. In the MCP section, open ChatGPT or Claude.
Copy the MCP server URL shown by Sparki and paste it into the assistant's custom remote MCP connector setup. When the assistant asks how to authorize, choose OAuth. If it asks for OAuth endpoints, copy the authorization and token URLs from the same Sparki setup screen.
After authorization, test the connection by asking the assistant to use Sparki before answering. For example: "Use Sparki to find recent unread inbox items that need follow-up."
Access and Privacy
The connection is scoped to the portal and user that authorized it. The assistant can only receive data returned by the approved Sparki tools and granted scopes. Sparki logs OAuth events and MCP tool calls so admins can review connected AI usage.
Use the Connected AIs section in the setup screen to review active connections. Revoke access when a connection is no longer needed, when a teammate changes role, or when you are testing with a temporary assistant account.
What the AI Should Not Do Alone
The assistant is useful for reading context, summarizing, drafting replies, and pointing your team to the right dashboard page. Write tools always require explicit approval. Review the exact customer, appointment, workflow, and proposed change before confirming it.
AI integrations are currently best treated as an alpha capability for careful pilots. Start with low-risk read-only use cases, such as looking up projects, summarizing recent interactions, or preparing a draft response. Be more cautious with write actions, because logged actions do not automatically mean every destructive change can be rolled back.
Good early use cases include answering customer questions, finding available properties, looking up recent customer history, preparing support notes, and helping a team member find the right dashboard page. Appointment cancellation should still be reviewed by a person before approval. Scheduling and rescheduling remain dashboard actions.
The MCP connector can send a portal message, cancel an appointment, update workflow metadata, and export customer records after explicit approval. Calling a write tool without approval returns a preview and does not change data. Appointment rescheduling and structural workflow edits remain dashboard actions.
Plan and Role Notes
AI add-ons are self-serve where available for the portal. Members need admin access to set up or revoke AI connections.