AI Integrations — MCP Server
G-Serwis provides an MCP (Model Context Protocol) server for workshop staff. Once a staff member is signed in, an AI assistant — ChatGPT, Claude, Cursor or Lovable — can read requests, orders, the fleet and the cash status, and change an order's status. Only staff accounts have access.
https://www.gserwis.com/mcp
How to connect
- 01
In your AI client, add an MCP server at https://www.gserwis.com/mcp (transport: Streamable HTTP).
- 02
The client will redirect you to the OAuth consent page — sign in with your G-Serwis staff account and approve access.
- 03
Once approved, the client fetches the tool list automatically. Access can be revoked in the panel: Admin → AI Agents.
Available tools
| Tool | Parameters | Purpose |
|---|---|---|
| workshop_summary | — | Business overview: fleet and availability, new requests, customer count, open orders and cash status. |
| list_service_requests | limit?: number (1–100, domyślnie 20) | Requests from the service form, newest first. |
| list_service_orders | status?: string, limit?: number | Workshop orders with customer, vehicle and value. |
| set_service_order_status | id: uuid, status: "new" | "in_progress" | "waiting_parts" | "done" | "cancelled" | Changes an order's status; status "done" stamps the closing time. |
| list_rental_bookings | status?: string, limit?: number | G RENT bookings together with the assigned car. |
| list_fleet | — | Full fleet with internal data: registration number, VIN, insurance and inspection due dates. |
Example call
Calls go through JSON-RPC 2.0 with an OAuth token in the Authorization header:
curl -X POST https://www.gserwis.com/mcp \
-H "Authorization: Bearer <token OAuth>" \
-H "Content-Type: application/json" \
-H "Accept: application/json, text/event-stream" \
-d '{
"jsonrpc": "2.0",
"id": 1,
"method": "tools/call",
"params": {
"name": "list_service_orders",
"arguments": { "status": "in_progress", "limit": 10 }
}
}'Security and limits
- Every call runs on behalf of the signed-in staff member — data is restricted by RLS and the is_staff function, not by the MCP server itself.
- A limit of 30 calls per minute per staff member protects against flooding; exceeding it returns a tool error.
- Every call is logged (who, tool, parameters, result, time) and visible in Admin → AI Agents.
- Tokens are never stored or returned in tool responses.