A common request is to analyze thousands of tickets at once. Asking your AI assistant to "read all my tickets" does not work well, because the AI assistant runs out of space to hold the data. This is a limit of the AI model, not of the Swifteq MCP Server. This article explains how to work at scale.
For standard support metrics, such as first reply time, resolution time and one-touch rate, you do not need any of this. Ask your AI assistant for a ticket metrics summary: the MCP Server calculates the figures from up to 3,000 tickets in one request. For SLA achievement, ask for an SLA summary, which covers up to 2,000 tickets in one request. Refer to MCP Server Tools and Limits.
In this Article
Two Things to Know
Recommended Approach: Extract, Store, then Report
Tips
Two Things to Know
- The 1,000 row limit only applies to CSV exports. Normal search results return 100 tickets at a time, and your AI assistant can keep requesting more pages. Ask it to "keep going until you have all of them". If you need a CSV of more than 1,000 tickets, split the export by date range and run the parts one after another.
- Retrieving tickets is not the same as analyzing them. Past a few weeks of tickets, the AI assistant cannot keep all the details in view at once. You need to reduce each ticket to the few facts you care about first.
Recommended Approach: Extract, Store, then Report
- Choose the fields you care about. Keep the list short, for example contact reason, sentiment, root cause, agent and whether the issue was resolved.
- Work in batches. Ask the AI assistant to go through tickets one date range at a time and, for each ticket, extract only those fields.
- Store the results in Zendesk. Ask the AI assistant to save the extracted fields on each ticket, as a tag or a custom field value, for example a Contact reason drop-down. The results stay in your own Zendesk account, with no other storage to manage. This needs write access, and updating tickets runs your Zendesk triggers, so first check which triggers react to tag or field changes.
- Report from the stored results. Ask the AI assistant to build the report from the tags or field values rather than the original tickets, for example by counting the tickets with each value. Because each result is small, it can cover far more tickets.
- Repeat regularly to keep the results up to date with new tickets.
In Claude, you can save steps 2 to 4 as a Skill. In ChatGPT, you can use a Project. We can share an example Claude Skill and ChatGPT Project setup on request. You will usually need to adjust the extraction instructions to suit your tickets.
Tips
- Use counts instead of searches whenever you only need numbers. One count can be split by status, group, channel, a drop-down field, CSAT rating, week or month, so ask for the whole breakdown at once.
- Ask only for the ticket fields you need. Descriptions and custom fields take up the most space.
- If accuracy matters, export the exact set of tickets to CSV first, then ask the AI assistant to report only on that export.
- Allow the read-only tools to run without asking for approval each time, so long analyses are not interrupted. In Claude, set Read-only tools to Always allow in the connector's settings.
For the tools and limits involved, refer to MCP Server Tools and Limits.
If you need help, please contact Swifteq support using the Live Chat or email support@swifteq.com.