Known Limitations of the MCP Server

This article lists the current limitations of the Swifteq MCP Server for Zendesk, so you know what to expect. Many of these come from the Zendesk API or from the AI assistants themselves.

In this Article
Help Center Articles
Macros
Data the MCP Server Cannot Access
Actions that are Not Available
Reports and Exports
Connection and Platform

Help Center Articles

• Updating a published article takes the live version offline. Updates are saved as drafts, and the Zendesk API cannot keep the published version live while a draft is saved. Publish the article again in Zendesk after reviewing it.
• Articles in Zendesk's new Knowledge editor can switch back to the previous editor when updated through the API. This is a Zendesk limitation.
• Articles cannot be deleted through the MCP Server. Delete them in Zendesk.

Refer to Replying to Tickets and Updating Help Center Articles with the MCP Server.

Macros

• Macros cannot be deleted through the MCP Server. Deactivate them instead: they can be reactivated at any time. Delete them in Zendesk if needed.
• Macros that start a side conversation cannot be applied. Apply them in Zendesk.
• Dynamic content, the translations a macro can use, can be read but not changed. Change it in Zendesk.
• Usage counts, such as how often a macro was used in the last 30 days, need Zendesk Professional or Suite Growth or above.
• Which tickets used which macro is not available.

Data the MCP Server Cannot Access

• Zendesk Explore reports and Help Center article views, which Zendesk does not provide through its API. Standard ticket metrics, such as first reply and resolution times, are available.
• Side conversations, followers, suspended tickets, views, and Community posts.
• Ticket change history (audits) as a separate tool, and linked Jira issues.
• Zendesk QA reviews and scores.
• The settings, flows and reports of Zendesk AI agents. The conversations they handle are available as tickets.
• Data held only in other apps connected to Zendesk, unless it is saved in the ticket.

Actions that are Not Available

• Deleting tickets, and marking tickets as spam.
• Merging tickets from different requesters. The MCP Server only merges tickets from the same requester.
• Adding followers to a ticket.
• Removing a file from a comment once it is posted. Files can be redacted in Zendesk.
• Attaching a file from your computer in an AI assistant that cannot run commands, such as Claude on the web without code execution. Files already in Zendesk can still be attached.
• Draft ticket replies, because Zendesk does not support them. Use internal notes instead.
• Creating or changing triggers, and changing Zendesk settings or forms.
• Editing users and organizations, and removing someone from an organization. The MCP Server can look them up and add people to organizations.
• Finding a user by the phone number field in their profile, and free-text search across users and organizations. Users can be found by email, name, role or tag, including a mobile number stored as a tag, and organizations by name or tag.
• Turning off individual tools on the MCP Server. Most AI assistants let you turn individual tools on or off in the connector settings.
• Taking a messaging conversation over from a Zendesk AI agent, or handing it back to the AI agent. Do this in the Agent Workspace.

Reports and Exports

• Ticket counts can differ from Zendesk Explore. The MCP Server counts tickets in real time, while Explore updates later. Use Explore as the reference for historical reporting.
• Split ticket counts cover up to 200 values per request, for example 200 groups or 200 days. Counts cannot be split by tag, organization, requester or assignee yet: the AI assistant counts each one separately.
• CSAT with customizable surveys: Zendesk records a neutral answer (3 out of 5) as a bad rating on the ticket, so it lowers the CSAT score. To see each answer's own rating, ask for the CSAT ratings themselves.
• CSV exports are limited to 1,000 rows. Split larger exports by date range.
• Ticket metrics summaries cover up to 3,000 tickets per request, and a day-by-day trend up to 31 days. Split longer periods. Averages and rates from separate requests can be combined, but medians cannot.
• SLA summaries and the SLA queue cover up to 2,000 tickets per request. Split longer periods into weeks, or narrow the queue by group or agent.
• SLA details of older tickets: Zendesk stops returning SLA details for tickets that are closed and not updated for about 120 days. SLA summaries leave those tickets out and say how many, so keep each month's SLA figures when the month closes.
• SLA figures for a past period can change when a ticket is reopened, because Zendesk can then turn an achieved target into a breached one.
• Group SLAs, which time how long a ticket stays with a group, are not included.
• Large requests run two at a time. Each Swifteq account runs at most two metrics summaries, SLA summaries, SLA queues and CSV exports at the same time. Ask for them one after another.
• CSV files can display badly in Excel when cells contain formatted text. Open them in Google Sheets, or ask your AI assistant to clean up the CSV first.
• Ticket due dates cannot be used as a search filter. Created and solved dates can.
• Searching and exporting tickets can fail on smaller Zendesk plans.
• Zendesk AI agent tickets can be listed and counted on messaging only. Search finds the tickets the AI agent resolved alone, the ones it escalated to a person, and the ones Zendesk counted as automated resolutions. Tickets an agent took over from the AI agent, and email AI agent tickets, are recognized one ticket at a time but cannot be searched, because Zendesk records nothing searchable for them. On email, Zendesk does not mark AI agent tickets, so the AI agent's replies are recognized by how Zendesk posts them.
• Ticket metrics and SLA reports cannot be split into AI agent and human tickets yet. Zendesk does not count the AI agent's messages as replies.
• Created dates of handed-over tickets. When a messaging AI agent hands a ticket to a person, Zendesk changes the ticket's created date to the time of the handover, so date filters select it by that time. The AI assistant can still see when the conversation started.

Connection and Platform

• Sessions expire and you need to sign in again. Refer to The MCP Server Connection Keeps Dropping.
• There are no API keys or long-lived tokens for unattended automation.
• Each Swifteq account connects to one Zendesk account.
• Gemini, Codex desktop, Open WebUI and Amazon Q Developer do not currently work. Refer to Requirements and Supported AI Assistants for the MCP Server.
• Each person adds the MCP Server to their own AI assistant. There is no way to add it for a whole organization at once.
• Swifteq accounts do not support SAML SSO, Google SSO or enforced MFA.
• The MCP Server cannot be self-hosted.

If a limitation is blocking your work, let us know. Requests help us decide what to build next. Contact Swifteq support using the Live Chat or email support@swifteq.com.