Connect your AI to Zendesk

What your AI can do with Zendesk

Once connected, these tools appear automatically in your AI assistant:

• search_ticketsread-only

  Search tickets with Zendesk's own search syntax, e.g. `status:open assignee:none`, `requester:a@b.com created>2026-01-01`, or plain words. This is the tool to reach for first — it answers most questions in one call.

• list_ticketsread-only

  List tickets, most recently updated first. Prefer search_tickets when filtering.

• get_ticketread-only

  Read one ticket. Use list_ticket_comments for the conversation itself.

• list_ticket_commentsread-only

  The ticket's conversation, oldest first. `public: true` is what the requester saw; `public: false` is an internal note only agents can read.

• create_ticketwrite · needs approval

  Open a ticket. The requester is an existing user id, or a name+email pair for someone not yet in Zendesk. Ids for group and assignee come from list_groups and search_users.

• update_ticketwrite · needs approval

  Change a ticket's status, priority, assignment or tags. To reply or leave a note, use add_ticket_comment instead.

• add_ticket_commentwrite · needs approval

  Add a comment to a ticket. It is an INTERNAL NOTE by default, visible only to agents. Set public to true only when the intent is to reply to the requester: Zendesk emails a public comment to them immediately, and it cannot be taken back.

• search_usersread-only

  Find agents and end users by name, email, or Zendesk search syntax such as `role:agent` or `organization:Acme`. Ids from here feed assigneeId and requesterId.

• get_userread-only

  Read one user (agent or end user) by id.

• list_organizationsread-only

  List organizations (the companies requesters belong to).

• get_organizationread-only

  Read one organization by id.

• list_groupsread-only

  Agent groups — the ids to route a ticket to.

• list_ticket_fieldsread-only

  Ticket fields, including this account's custom ones and their allowed values — what a ticket can actually carry here.

• list_viewsread-only

  Saved views — how this team slices its queues.

• list_macrosread-only

  Active macros — the team's approved canned responses. Read-only here: use one as the basis for a reply rather than inventing wording.

• list_brandsread-only

  Brands configured on the account (multi-brand help centres and addresses).

• search_articlesread-only

  Full-text search across the help centre.

• get_articleread-only

  Read one help-centre article, with its body.

• list_sectionsread-only

  Help-centre sections — where an article can be filed.

• create_articlewrite · needs approval

  Write a help-centre article. It is created as a DRAFT: a person publishes it from Zendesk. sectionId comes from list_sections.

• update_articlewrite · needs approval

  Update an existing help-centre article's title or body. Its published state is left exactly as it was — this never publishes and never unpublishes.

• list_satisfaction_ratingsread-only

  Satisfaction ratings left by requesters, most recent first.

Example prompts

Just ask your AI in plain language — it picks the right Zendesk tool:

• “Search tickets with Zendesk's own search syntax, e.g. `status:open assignee:none`, `requester:a@b.com created>2026-01-01`, or plain words. This is the tool to reach for first — it answers most questions in one call — from Zendesk, in plain language.”
• “List tickets, most recently updated first. Prefer search_tickets when filtering — from Zendesk, in plain language.”
• “Read one ticket. Use list_ticket_comments for the conversation itself — from Zendesk, in plain language.”
• “The ticket's conversation, oldest first. `public: true` is what the requester saw; `public: false` is an internal note only agents can read — from Zendesk, in plain language.”

How it works

1. Create your Proxoria account and add your MCP endpoint to Claude or ChatGPT.
2. Connect Zendesk with a secure OAuth connection — each user connects their own account, credentials encrypted at rest.
3. Your AI instantly gets the Zendesk tools. Write actions wait for your approval in the dashboard before anything changes.

Frequently asked questions

Does Zendesk have an official MCP server?

You don't need one. Proxoria exposes Zendesk to your AI through the Model Context Protocol, so you can use it from Claude, ChatGPT, Mistral or any MCP-compatible client — no official Zendesk MCP server required.

What can my AI do with Zendesk through Proxoria?

It gets 22 tools: search_tickets, list_tickets, get_ticket, list_ticket_comments, create_ticket, update_ticket, add_ticket_comment, search_users, get_user, list_organizations, get_organization, list_groups, list_ticket_fields, list_views, list_macros, list_brands, search_articles, get_article, list_sections, create_article, update_article, list_satisfaction_ratings. For example, search tickets with zendesk's own search syntax, e.g. `status:open assignee:none`, `requester:a@b.com created>2026-01-01`, or plain words. this is the tool to reach for first — it answers most questions in one call.

Is the Zendesk connector read-only?

No — it can also make changes. Write actions go through Proxoria's confirmation guardrail by default: your AI proposes them, and you approve each one from your dashboard before anything is written to Zendesk.

How do I connect Zendesk to my AI?

Create a Proxoria account, connect Zendesk with a secure OAuth connection, and add your Proxoria MCP endpoint to Claude or ChatGPT. Your AI instantly gains the Zendesk tools.